31 December 2013

End of year 2013

Seeing as its new years eve I thought I would give a final update with some statistics. As you can see from the BSrac (BOINC Stats recent average credit) column I have been concentrating on Asteroids.

For next year I'd like (weather permitting) to be able to concentrate on GPUgrid and if they ever come back Orbit followed by the others. I figure that GPUgrid can help solve medical mysteries and Orbit can help search for near earth objects that could potentially wipe out the planet.

Future plans
As always I keep an eye out for GPU upgrades if they make sense. That is they need to provide a performance gain without using more electricity than my GTX670's (which use 185 watts max).

In the low-power computing area I have a couple of Parallella's coming. Maybe, one day. There should also be more multi-core ARM based machines out next year. We will see.

29 December 2013

29th of December

Farm news
We've had a couple of cool days during the week. I'm currently on 9.6 million and trying to get to 10 million credits for Asteroids. Crunching continues on the Intel GPU's for Einstein as well.

I consider over 7 days to complete an Asteroids work unit a bit too extreme. We asked for smaller work units for slower computers but they weren't willing to have multiple size work units. There is a possibility that may change when they get a GPU app working. Until then I have reallocated all 5 of the Raspberry Pi's to Einstein.

Parallella news
We got another update that they have shipped approximately 130 units so far.

Here is a thermal image of the Parallella. As you can see the Zinq 7010 chip gets rather hot.

This is how it looks when they added a heatsink and fan to provide air flow. As you can see somewhat cooler. They have decided to provide a heatsink with each Parallella for now.

The black chip to the right is the Epiphany which wasn't in use at the time. According to reports from them it too gets rather hot and will also need a heatsink.

22 December 2013

22nd of December

Farm news
We're still off all this week.
I got around to replacing the couple of switches below as I have used all the ports and they don't make these stackable ones any more. So we went from these two 8 port switches
To this 24 port switch. It looks messier because of the cabling but its actually one less than before. I need to tie them off and re-route the power cable but its working at the moment. One advantage of this switch (apart from the extra ports) is its rack mountable.

While I was at it the router also died. Maybe as a result of swapping out the two switches although it should make no difference. I ended up buying a new one. They only seem to last a couple of years. I had to spend a bit of time configuring things to get it working.

12 December 2013

12th of December

Farm news
Not much to report. Its been hot so everything is off, except for the Raspberry Pi's.

Parallella news
They got their first batch of bare boards back and they are a blue colour. These were promptly sent off to have the components fitted and the first batch are back and being tested. They expect the initial 200 boards will be shipped before Christmas. The rest will follow as they are completed. I am hoping I will get mine before Australia Day (26th of January).

Project news - Asteroids
They have doubled their work unit size again. This isn't much of a problem for my larger crunchers but the Raspberry Pi's are now taking over 7 days per work unit. Hopefully we can have smaller work units for the slower crunchers otherwise I will have to switch my Pi's over to Einstein (which does have smaller work units for the slower crunchers).

It now takes around 80 work units to calculate an asteroids shape.

04 December 2013

4th of December

Farm news
On the weekend it cooled down for a while so I had the GPUgrid crunchers running for a couple of days until things got too hot again. I had one work unit that refused to upload so I had to dig out the trusty backup file server and hook it up to the dialup modem in order to get the file uploaded. It took about an hour to get though it.

I also installed the additional 2Tb hard disk in the file server. This gets rather time consuming because of the backup and restore. The installation was fairly straight forward apart from the machine refusing to boot. I tracked that down to a loose SATA cable on one of the other hard disks. The file server now has a 5.45Tb array made up of 4 x 2Tb hard disks in a RAID configuration. I have used up all its on-board SATA ports and I don't think it will support individual drives larger than 2.2Tb due to its age.

BOINC testing
We got 7.2.33 which was also the release version last week. The main change since .31 was some more code around preventing images being displayed in notices.

Parallella news
They provided an update. Once again the expected delivery dates have slipped. We're now looking at the end of December before they are likely to arrive. They had some non-technical issue with the manufacturing which was contracted out.

There is supposed to be an announcement today from them but no indication as to what it will be about.

Project news - Asteroids
They went off line a couple of times last week due to server upgrades. They are off-line again for the next couple of days for another server upgrade, however the sysadmin forgot to warn us before.

Project news - Einstein
They've discovered four "young" gamma ray pulsars. These are new discoveries. You can read the news item which has links to the scientific journal and a press release HERE

23 November 2013

23rd of November

Farm news
During the week I rebuilt the old file server. Its now back to its former self with the RAID controller deciding to work, although I did have to work out how to setup the RAID all over again. I have removed the 56k modem for the time being.

This weekend I picked up a new KVM to replace the broken one. That meant I could now access the machines that it services. They've been off for about a month. There were windows updates, BOINC updates, Nvidia driver updates and even SSD software updates. Repeat that for 5 of them so that took a while.

While I was at the computer shop I also picked up another 2Tb hard drive. Its destined to go into the file server. I'll probably do it tomorrow.

BOINC testing
7.2.28 was release to the public last week. This week they found a problem with notices containing images for one particular project. There was a quick update to disable messages and this version (.31) is now the recommended one.

The Raspberry Pi's still have 7.0.27 in the Debian repo at the moment. 7.2.28 is in the experimental releases so don't expect it to make it into the main repo any time soon.

13 November 2013

12th of November

Farm news
The Intel-GPU part of the farm has been busy doing Asteroids and Einstein BRP4 work.

CPDN uploads have stopped for the moment as their server has run out of disk space. I still have 2 machines left to clear via the 56k dial-up.

I spent about 4 and a half hours on Saturday night swapping out the hard disks for larger ones. The process of physically removing the old ones and add new ones was easy enough. After realising one has to define the RAID to the BIOS and watching the boot up messages a few times to find out what the hotkey was, it was up and running. Then I found out I had a 2010 version of the Intel Rapid Storage Technology software that only supported a RAID with a total size of 2Tb. I go to the Intel site and download the up to date version. It didn't install properly. It was a bunch of INF files and a driver. After stuffing around for about 30 minutes I gave up with the Intel software and decide to partition the drives via Windows. I picked the wrong boot sector method as it defaults to MBR, which only supports 2Tb so I end up with 2 logical drives. Back into the BIOS delete the RAID, reallocate the RAID and then back into Windows and do it right this time. I then go back to the Intel site and find a setup program for the Rapid Storage Technology software and give that a go. That fixes the Intel software and I can now manage the RAID. After that I restore the backup which went through without issue, but by the time it was all done it was 2am.

BOINC testing
We got a 7.2.28 which was also released to the public this week. It was really just a change to the installer so no real change to the software. I recommend people upgrade to this one as it adds some new features since the last release (7.0.64/65) and fixes some bugs in the process.

03 November 2013

3rd of November

Farm news
Some good news this week. Firstly the replacement power supply for the Drobo arrived, so it was pressed into service the day it arrived.

Also I picked up the 2Tb hard disks which I will be putting into the file server soon. I needed the Drobo to backup the existing drives first.

The older file server is now an additional proxy server, the one I mentioned rebuilding last week. It and the dial-up have been going all week to upload CPDN tasks. I have cleared 2 out of 5 machines so far but its very slow going. Files take 10 hours to push through. I had a wasted day where the wife kept trying to use the phone and that meant the 56k modem kept being disconnected.

I am still waiting on a replacement KVM so I can't access half the farm. That's the half with the Nvidia GPU's, but then its been too hot to use them anyway.

BOINC testing
Just after I wrote last weeks post we got 7.2.26 which is expected to be released on Monday as a the "production" client. I have installed it on the few machines that were running a slightly older version. Everything seems to work as expected.

Project news - Asteroids
They've release some details of work we've been crunching. We have managed to calculate the shapes of 114 more asteroids which brings the project up to 390 so far. The link to their results is http://asteroidsathome.net/scientific_results.html

26 October 2013

25th of October

Farm news
More bad news this week. The KVM that I mentioned last week has been pronounced dead as a dodo. I have had to order a new one to replace it. In addition the stock they expected to get didn't come in so its been back ordered and is now expected in late November.

The power brick for the Drobo is dead as a dodo as well. The good thing about it is its under warranty and Drobo have already shipped a replacement out of Singapore. Its expected early next week. Hopefully the Drobo itself will be fine.

Last week I went to update the old file server only to find its LAN card had also died. It hasn't even been plugged in for 6 months. No problem I had an old 100Mbit one brand new in a box, so I swapped out the old one. Then the machine wouldn't boot and I had problems getting it to work. I removed the RAID controller and spent the rest of the day reinstalling Windows. In the end I put an 80Gb hard disk in it that I also had brand new still sealed in its anti-static bag. I reinstalled windows yet again and then left it to download the 170 updates which took all day to get and install.

The 56k dial-up is still waiting on the ISP to give me account details. I haven't been able to start using it until they can provide this information. My current thinking is to use the old file server as a proxy server with the 56k dial-up and then switch each machine to using it as needed.

My CPDN work is still failing to upload and scheduler requests are still failing. Its been about 3 weeks now.

I have to pickup a bunch of 2Tb hard disks from the shop tomorrow. These are destined to replace the 1Tb drives in the main file server. This is more a preventative maintenance thing as well as doubling the available space.

BOINC testing
we're up to 7.2.25 this week. It has a correction around Intel GPU exclusions and a change to the way VirtualBox memory limits are set. I haven't tried it yet but I do have one machine which has been running 7.2.23 for almost a week so I will update it.

Parallella update
They missed their October delivery deadline. One of their key staff left. As well as that one of their suppliers decided to end-of-life one of their components so they had to rework the board design. The good news is they have replaced the key staff member and rejigged the board. They also made the Epiphany III chip available for public sale (ie without the Parallella) after they have shipped the Parallella's to backers and pre-orders.

15 October 2013

15th of October

Farm news
We had a few hot days again this week where the farm was completely off.

Late last week I did a UPS test by switching off the power board that its plugged into to see if the server continued running. It did. After 10 minutes I turned it back on to find the KVM was no longer working. After a number of attempts to get it going I have had to take it back to the shop. New ones cost $600 and the distributor doesn't have stock until next month. I have asked to get it repaired and of course its not under warranty. Because the other cluster of machines are off I have swapped the other KVM over to get going.

The Intel-GPU machines are still finishing off their climate models. The Raspberry Pi's are running Einstein work because Asteroid ran out of work.

BOINC testing
We got up to 7.2.21 early this week (or .22 on the Mac). The main change with this is OpenCL support of CPUs.

Project news - Climate Prediction
I still can't upload. This week their server is reporting its out of disk space. As well as that I can't report any completed work.

I have ordered a 56k dial-up modem which I will use via another ISP in case of issues like this.

08 October 2013

8th of October

Farm news
On and Off is probably the best way to describe the past week. We had a few hot days so the farm was off for those and a few cooler days where the Intel GPU machines were running. The farm is still concentrating on CPDN work with their 240+ hour full resolution ocean (RAPIT) work units.

While that's going on I am running the Einstein BRP4 work units on the Intel GPUs. They take around 13 minutes a work unit so between 5 machines they can get a fair few completed in a day.

The Raspberry Pi's are also happily running a mix of Einstein BRP4U and Asteroids work units.

Climate Prediction woes
Its been a rather frustrating two weeks. I can't upload most of the CPDN results. The uploads are failing part of the way through. I have tried using UK-based proxy servers and no proxy server but it seems to make no difference. I even went back to an older version of the BOINC client just to prove that it wasn't the alpha-test version I use. Some of the step files have tried 26 times to upload.

To make matters worse I get "HTTP Internal Server error" messages when trying to report the few work units that have completed. Two work units have failed and a few have managed to upload their result files. You get 4 step files per work unit - if they work. I won't be doing any more CPDN work any time soon.

30 September 2013

30th of September

Farm news
Its getting hot! I've had to reduce crunching to overnight at the moment, with the occasional all-day run when it stays cool for a day or so.

At the moment the farm is concentrating on CPDN RAPIT (full resolution ocean) tasks. They take around 10 days each. The problem with running these is they exclude everything except GPU work from running. To make matters worse I get errors while trying to upload their step files (which are approx. 52Mb each). I have a couple that failed but I can't report them either.

Project news - Asteroids
They finally came back online on Thursday last week. Their was then a lot of contention to upload completed work seeing as everybody else was trying to do the same. The back log of work getting uploaded was so great they had to run additional instances of their validator to get through all the uploaded results.

This week they released revised optimised apps which are a bit faster again and should resolve a bug that prevents completed tasks from validating. The project admins have indicated they will be looking to increase the size of work units now they have faster apps.

21 September 2013

21st of September

Farm news
No too much happening on the farm recently. As the weather has warmed up the Intel-GPU machines and the Raspberry Pi's are all that's running. They are running a mix of Asteroids and Einstein work, well they have been up until yesterday when they finished off all the Asteroids work.

I am still waiting on Parallella news, hopefully shipping confirmation. I suspect it will be October before hearing anything, which was their worst-case date for shipping to KickStarter backers.

Project news - Asteroids
They were recovering from their server crash last week when they had a power failure which took them off line for a day. They have been off-line with what was supposed to be a scheduled upgrade to a new server. It was estimated to take 8 hours but it has been off-line for more than 24 hours so far.

Project news - ClimatePrediction
They seem to have finally recovered from their failed server that were going to retire. They released a bunch of RAPIT (Full resolution ocean) models. They then suspended that lot and released a bunch more that the scientists wanted back sooner. All my Intel-GPU machines have picked up 7 each, so they will be busy for the next 10 days.

08 September 2013

9th of September

Farm news
We had a warm weekend so I took the opportunity to clean some of the rigs. These are a couple of the Fractal designs ARC Mini cases that have ASUS P8Z77 motherboards. They have a metal grill with foam inside which traps the dust coming into the case so one has to remove them and wash under a tap.

Without the front panel. Its naked!

And the front panels

Project news - Asteroids
They are still having issues from their server failure last week. While they have got things going most of the files for the work units are missing so all the downloads are failing, It takes some hours running a script to keep asking the project for work before one can actually get them.

They released an updated optimised app today as well, which isn't the best timing. It would have been better to wait until the download issues are resolved. It seems to be about one third faster than the previous one.

03 September 2013

3rd of September

Farm news
Its started warming up now so I have shut down the GPU crunchers. That leaves the Intel GPU machines and the Raspberry Pi's running.

I received my confirmation email from Adapteva confirming I was getting 2 Parallella boards but only one case. I have now paid for an additional case. I am waiting on further news from them at the end of this week which will probably just be an order confirmation.

I ordered some USB to 2.1mm barrel connector cables off eBay. The US power packs that Adapteva sell won't fit into an Australian power point. I have a number of 1 amp USB power packs now (after moving the Raspberry Pi's onto dual-USB power packs) so with these cables I should be "Parallella Ready".

Asteroids project news
Their server crashed and they were off-line for a few days this week. They are back up now and have had to put their database on another machine to reduce load on their main server. They have another server on order so there will be a scheduled outage at some point when they are ready to switch over.

Despite the server being off-line I have managed to climb up to 9th place on the rankings for the project. I expect I will be dropping soon because they have large numbers of download failures.

I have over 900 work units fail to download in the last day. These ones don't even start due to missing files. The project has set the error count to 20 so it will try and send the same work unit to 20 different people before marking the work unit as failed. The BOINC client thinks it has a comms error so it backs-off up to 24 hours and stops communicating with the project. I will have to switch to another project for a few days while they sort out this issue.

25 August 2013

25th of August

Farm news
I got the KVM cable on the weekend so the new i7 is off and running now. I spent a bit of time on Saturday.setting up projects applying windows updates, etc.

I also got a couple more of the Belkin dual-USB power adapters during the week. I have removed the individual ones so the power board they were plugging into now has a few free sockets. I also tidied up the cabling a bit with a few cable ties. They all running quite happily at the moment.

The Seti WOW challenge is still running so I am staying away from Seti for the moment except for the one machine that has dual GTX660's.

Project news - Asteroids
Last week they released an SSE3 optimised app. This week they released an SSE2 one for older computers. The scheduler has been sending these out so I've been getting a mixture of SSE2 and SSE3 apps. After about 10 results the scheduler can determine which is the fastest and only sends that app. There is about 1 minute difference between the two apps when running on my i7 machines. They take about 30 minutes a work unit. The project admin has mentioned they will look at increasing the work unit size later.

19 August 2013

19th of August

Farm news
Crunching has shifted to Asteroids for the moment due to the Seti challenge that I wrote about last week. Work requests aren't getting much work so I have left a dual-GPU machine running it while the others run Asteroids.

Belkin make a dual USB power supply that is rated for 2.1 amps per USB port, more than enough to run a Raspberry Pi or Parallella. The sales rep in the shop was rather worried that it would only work for those products listed on the box (Apple products) and tried to talk me out of getting it. It seems fine running one Raspberry Pi at the moment. I will plug another one in and see how it goes but I am not expecting any problems. Assuming it works out I will get a couple more to free up some power points.

Still waiting on the KVM cable to get the newest i7 going. Hopefully it will arrive during the week.

Also still waiting on the Parallella boards to arrive. There hasn't been any news about them shipping yet but here's hoping.

BOINC testing
We got to 7.2.11 this week. No major changes, just some minor tweaks around notices and the rest is mostly Mac related. I have installed it on the iGPU machines and haven't noticed any issues so far.

Project news - Asteroids
I mentioned they released optimised apps last week. This week they have released an SSE3 version of their app which is taking about half the time (approximately 35 mins) that the optimised app takes per work unit. There is still an AVX version to come but that is waiting on some changes to the BOINC client.

15 August 2013

14th of August

Farm News
The farm continues to concentrate on Seti work for the moment.

I picked up new i7 on the weekend and installed stuff on it. Unfortunately I don't have a long enough KVM cable for it, in its new location. I have ordered another (longer) one so I can get it hooked up.

I had to reimage one of the Raspberry Pi's as it wouldn't let me access it except by plugging in a keyboard, mouse and monitor. That fixed the access issue but the image I have is from April and so I had to apply a fair few updates after that.

I have started getting some bits for the Parallella's. I'm not sure when they will arrive but thought I should start getting ready. I needed a couple of USB to Micro-USB adapters so I could plug in a keyboard and mouse. I also got a MicroSD card for it. I still need to get a power adapter and might rearrange the ones running the Raspberry Pi's. I'll probably get a few dual-USB power packs otherwise I will run out of power points.

The Orbit@home project is looking to restart in "the Fall of 2013" to quote their words. They have received some funding from NASA.

Project News - Asteroids
They release optimised apps for all platforms. The Raspberry Pi's had about a 1/3rd speed up. The Windows and Linux users are reporting a 50% speed up. There are even further optimisations in the works (ie using the AVX and SSE3 instruction sets) which are still being tested.

Project News - Einstein
They've moved the BRP4U app from the beta test project (Albert) to their main project, so when the Raspberry Pi's stopped getting work I attached them to Einstein.

They still have the BRP4 work units running on the Intel HD graphics using an OpenCL app.

Their beta test project is also testing a new content management system (Drupal) so they have reworked all the web pages. It takes a bit of getting used to.

08 August 2013

7th of August

Farm news
I did some rearranging of the network on the weekend. Unfortunately it went wrong and I was without internet connectivity for a couple of days. Well a couple of machines could connect, just not all of them. I think I have things working again, but it will take a couple of days to confirm as the slower machines finish off work.

I should be picking up another i7 on the weekend which will join the Intel-GPU cluster once set up.

Android BOINC
As I mentioned last week various projects are supporting the Android platform. That has generated a lot of messages on the various project message boards about their problems. Personally I have no plans to use an Android based phone or tablet for number crunching as I don't consider them suitable.

Seti Wow! Event Challenge
SETI.Germany is organizing a challenge to celebrate and honour the 36th Anniversary of the Wow! Signal detection.

The challenge will run from 16:00 UTC on the 15th of August to 16:00 UTC on the 29th of August. If you are interested see the SETI.Germany web site HERE

27 July 2013

27th of July

Farm news
The farm continues to process Seti work for the moment. A couple of the Intel-GPU machines are also running the Einstein BRP4 OpenCL app.

This morning I dropped off the P8Z77-V LE Plus motherboard at the shop along with a hard disk and DVD burner. They will be supplying the other bits for this build. This will add another machine into the Intel-GPU cluster.

BOINC testing
I spent a bit of time running 7.2.5 alongside the 7.1.18 and haven't seen any difference in behaviour so far. I have updated a few more machines to 7.2.5 as it seems fairly stable.

BOINC for Android
The Android version has been officially released via the Google Playstore and the Amazon appstore (for the Kindle Fire). You'll need Android 2.3 or later. The following projects have apps for it:
  • Asteroids@home
  • Einstein@Home
  • O-Project@Home
  • theSkynet POGS
  • World Community Grid
  • Yoyo@Home
Check the individual project forums for the minimum requirements (memory mostly).

Parallella news
This is the Parallella Cluster Kit that is now being offered

See http://shop.adapteva.com/ for details. They are now also accepting pre-orders for the Parallella-16 (and the Cluster Kits) for October shipping.
And this is a 42 Parallella cluster that they setup with a total power consumption less than 500 watts. For the full details head over to www.parallella.org

21 July 2013

21st of July

Farm news
This week continued concentrating on Seti with the Einstein BRP4 app on the Intel GPU's. There was a bit of warm spell during the week so I had to stop crunching on the Nvidia GPU machines and its the middle of winter here in Sydney, Australia.

We got a bunch of Windows updates last week which I finally got around to applying to all the machines this weekend. I also took the opportunity to update the Nvidia drivers to the latest WHQL version. Oh and there was a VirtualBox update as well.

Hardware-wise the shop told me they couldn't get the ASUS P8Z77-V motherboards anymore, so I promptly bought one online. Although to be fair the place I got it from also told me they are discontinued. Hopefully that will arrive next week and the other bits can make their way to the shop for assembly the week after.

I am considering getting a couple of GTX770's to replace my 670's, but I am not in a hurry.

BOINC testing
We got 7.2.5 during the week. The main change is in the way GPU detection is handled now. There have been a couple of reports of it not working properly. Most of the farm is running 7.1.18 but I have 7.2.5 on one machine which appears to behaving. There have been a couple of unreleased versions (presumably for the Android) since 7.2.5 came out.

13 July 2013

13th of July

Farm news
This morning I had the file server off at the shop. While not one of the number crunchers it serves the whole farm.

It was in just for a clean. It was starting to get fairly dusty inside, the CPU heat sink and fan in particular. After killing two KFSN4 machines I don't touch the CPU and would rather pay the shop to handle it. The various other bits I clean myself. Anyway its back to normal duties now.

Einstein supports iGPU
They were testing an app for the Intel iGPU's. That is the HD2500 and HD4000 series of built-in GPU's that are on recent Intel CPU's. As there weren't many computers with the required hardware attached to their beta-test project (Albert@home) they decided to release it on the main project.

All my recent machines have iGPU's in them, although I have them disabled on the machines with other graphics cards. That meant the 4 machines that I have under the Intel-GPU cluster were all able to pick up work. It wasn't all plain sailing. It turns out there was a work-fetch bug that provided too much work (there still is in my opinion). Also I had to download the latest drivers from the Intel web site as the first lot of work unit kept failing. However once over these hurdles they have been going flat out for the last couple of days.

Here is the link to the Intel Download Center for the latest iGPU drivers:
Intel Download Center

Because they work so well I have ordered another machine for my Intel-GPU cluster. They take between 12 and 14 minutes to complete and they require a CPU core to be free.

03 July 2013

3rd of July

Farm news
Since my last post I have managed to finish off those CPDN RAPIT or "full-res ocean" work units. They only take about 210-220 hours each on the fast machines.

I also put two of the WD Blue drives into the machines that were running the left over drives from the Core2Quad machines. Looking at one of the old hard disks it says it was made in February 2008. They still work but its safer to replace them before they become a problem.

Now that I have cleared the CPDN work I have the Intel-GPU cluster concentrating on Seti as its total has dropped below that of some of my other projects. The idea is to run it for a while to help bring the points up to a similar level as Einstein. I still have other machines doing Einstein and Asteroids work, just not the Intel-GPU cluster.

They have release their hardware design as open source. They will in a short while be ordering mass production (about 6300 unit) unless there are some last-minute design changes.

There has been some interest in porting the Seti multi-beam app to run on the Parallella, not just from the usual optimised app developers but also from the Parallella design team. The idea is they can show it running an OpenCL app (Seti multi-beam) as a demonstration of what it can do. There is also a fair bit of interest from the crunching community in using them as low-cost computers.

It seems the final design has 1Gb of DDR3 RAM, a dual core ARM A9 processor, a 16 or 64 core Epiphany chip, 1Gbit LAN, 2 x USB 2.0 ports, HDMI connector and a MicroSD card socket.

24 June 2013

24th of June

Farm news
Yoda came back from the shop "no fault found". This is what it looked like when I ran memtest86+ v4.0a on it. Its overwriting the video memory with the test pattern.

After downloading the latest version (v4.20) it behaved. I left it running for 24 hours and didn't find anything. Its back to processing, just not Einstein work at the moment until they finish off the Gravity Wave LineVeto search.

The replacement monitor arrived and has been put into service.

The WD Blue hard disks also arrived (the same day as the monitor). I will be putting a couple of them into the Intel-GPU machines that are running old hard disks.

I allowed CPDN to download some RAPIT work units, unfortunately most of the machines got 7 of them. These climate models have been running for over 200 hours so far and still haven't finished. This includes the couple of machines I wanted to replace hard disks. Note to self: Don't select CPDN RAPIT tasks in future, they take too long!

BOINC testing
We got up to 7.1.18 this weekend. It has a few work-fetch tweaks and some stuff around not reporting work if uploading. It still seems to have issues with work fetch if using app_config restrictions.

16 June 2013

16th of June

BOINC news
We got 7.1.15 earlier in the week, it has since been superseded by 7.1.17. It contains mostly scheduling fixes. I am waiting on another build as there is a fix for one problem I reported to do with making unnecessary scheduler RPC calls.

There is also an Android version available via the Google Play Store. I don't have any Android devices so haven't seen it.

Farm news
I took Yoda back to the shop for them to work out what is happening with the memory failures. I suspect it will be the motherboard as I have never had a CPU fail, nor have I had issues with Kingston memory.

Last week I updated the Intel-GPU machines with the latest Seti optimised apps. This week I updated one of the Nvidia-GPU machines and allowed it to resume Seti processing. All working as expected.

I have 3 more WD Blue hard disks coming in to replace the various drives in the Intel-GPU cluster. They all were given recycled drives out of the old Core2quad machines I had before. I have swapped out 2 of them but the remaining ones need swapping out. Hopefully they will arrive tomorrow.

The replacement monitor should also arrive tomorrow, so I can at least log on to the Intel-GPU machines. While I can control BOINC on them I can't do any maintenance without moving the working monitor around.

Raspberry Pi news
Daniel Carrion has included the Multi-beam v6 app in his Seti build. When he updated before he included only the Multi-beam v7 app. If you need both (while they run off the v6 work units) you'll need to download it again. He has also completed a v7 work unit in 10 days. You'll need to run a minimal cache as the work units have a 14 day deadline.

Daniel's blog can be found HERE.

10 June 2013

10th of June

Farm news
It seems Yoda has some serious memory problems. I left memtest86 running and after a few hours it starts writing over video memory. I replaced the memory with some old Corsair 1Gb kits I had and it rebooted a couple of times during memtest. That suggest the motherboard or the CPU are faulty. So it will be off to the shop later this week.

I replaced the hard disk in it but that made no difference. I already suspected it was the memory but it takes a few hours running memtest before anything strange shows up. While I was at it I found a BIOS update so applied it. Of course that didn't make any difference.

I also grabbed a copy of the optimised Seti apps and installed them on all the Intel-GPU machines. I still have a dual GTX660 machine that needs to be updated however the monitor is over on the Intel GPU machines (waiting on a replacement for the dead 17" one to be delivered). Yes I can swap it around but its not urgent.

Raspberry Pi
Daniel Carrion has updated the apps for the Raspberry Pi so you can also get a Seti v7 multi-beam app for them now. See his blog HERE to get the files. A big thank you once again to Daniel.

They have been a bit delayed again. Their current estimate for the early access users (approx. 100) will be the end of June and for the rest of us backers (approx. 6300) will be the end of August.

For those that don't know what a Parallella is, its a small computer very similar to the Raspberry Pi in size. It has a dual core ARM processor along with an Epiphany chip which is OpenCL capable. There are currently 16 core and 64 core Epiphany chips.

If you missed out on backing the project you can still order one - head over to www.parallella.org and you can pre-register for when they become available to the public.

05 June 2013

5th of June

Farm news
Had a couple of weird things happening with Einstein work on one Intel-GPU host. It was failing a lot of the Lineveto work units.

I thought it might be memory issues so I took it off line and had it running Memtest86 for four hours which didn't find anything. I did however discover the DVD reader in it had failed, so had to replace that. I also ran a disk check which didn't find anything wrong. After bringing it back online I found it only happens when it tries to run 8 at a time. 7 at a time work fine. So I restricted it to run only 4 at a time using an app_config file.

The 17" LCD monitor that I was using on the KVM for the Intel GPU machines decided to die as well. I have ordered a 22" one from Dell. It was 10 years old.

I had a GPUgrid cruncher which failed one (and only 1) as well but it wasn't running 8 of them. That suggests to me they may be faulty work units but other users seem able to process them.

I recycled 4 Acer Core2Quad machines before Christmas and used the DVD and hard disks in the Intel GPU machines. I have purchased a couple of replacement DVD readers. The hard disks which also came out of the Acer machines are likely to get replaced as well. I have a couple of WD Blue drives as spares which will now get swapped in. I need a few more in order to do all 4 Intel GPU machines.

BOINC testing
We're testing 7.1.x at the moment. We got 7.1.3 this week which has some scheduler improvements around excluded GPU's. I put it onto the machine throwing all the errors for Einstein and so far haven't had any issues.

Seti Multi-beam v7
The optimisers have released a new windows installer for those that don't know how to do their own app_info file. It contains all the necessary updated apps. They had a problem hosting them on the Lunatics website so please use either of the links below:

  • Crunchers Anonymous site HERE (in the US)
  • Mikesworld site HERE (in Germany)

I need to download them myself so I can update my machines. I prefer to edit my own app_info file as the one created by the installer tends to have a lot of unneeded entries (so you don't lose any existing work if you have it).

30 May 2013

30th of May

A quick update from my last post...

Seti did indeed release the long-awaited multi-beam v7 app. This is causing a bit of confusion on what to do to upgrade, if at all.

If you don't use an app_info then you don't need to do anything. The project will simply supply the apps itself. As you can imagine there are a lot of people trying to download them now so a lot of  downloads are backing off.

The CPU and OpenCL apps are not yet available from a 3rd party site. The "stock" apps are your best option for now. I would suggest you finish off any multi-beam work you have, upload and report it. After that rename app_info.xml, restart BOINC and let it get the "stock" apps. Even better if you can delay this for as long as possible due to the load on the download server.

They too release their BRP5 work units that are much larger than the BRP4 work units. As there are no optimised apps for Einstein it should simply update itself and start work. I have managed to process a few work units already on my machine with dual GTX660 cards.

26 May 2013

26th of May

A word of warning
Next week Seti@home should be upgrading to the v7 multi-beam application. Its planned for, but then its been in planning for a long time so don't hold your breath, Wednesday US time. If you run the "stock" applications you don't need to do a thing. If however you run optimised applications you will.

I am in the group that runs optimised apps, so I am setting my machines to No New Tasks now so they finish off what they've got. The Cuda multi-beam x41zc app from Jason is "v7-ready" as they say so I don't need to do anything there, but the CPU and OpenCL ones need updating. Next week once things settle down I will look at downloading the latest apps. The guys from Lunatics are hoping to have a windows installer put together, but it won't happen right away.

Some useful links:
Jasons site for CUDA apps
Lunatics site for CPU and OpenCL apps

Project news - Einstein
In other news Einstein have released their BRP5 (Binary Radio Pulsar v5) app to replace the BRP4. The BRP5 work units take approximately 10 times as long as BRP4 and are GPU only. Unfortunately they aren't running the validator yet as they want to check it when it first runs, that should be early next week.

22 May 2013

Pi Pics

Chris asked to see some pictures of the Pi's. This is what they look like. Not a very good way of handling them, which is why I am searching for a suitable case.
Here is a close up of a few of them, with the top off one case so you can see what they look like.

This is the Simple Pi Rack II that I mentioned previously. Its the best multi-Pi case I have seen so far. There are lots of single Pi cases but hardly any for multiple Pi's. This one has power (via a 7 port USB hub) and a network switch underneath. The link to the message thread over in the Raspberry Pi forums is: http://www.raspberrypi.org/phpBB3/viewtopic.php?f=40&t=41669

19 May 2013

19th of May

BOINC Pentathlon
As I mentioned last week Asteroids@home were one of the projects being covered in the BOINC Pentathlon. They were the 5th project being run from the 14th to the 19th of May.

Even though I don't qualify as a team I had the farm working away on it. In fact they are still trying to finish off work. They should complete these later tonight even though the pentathlon has finished.

I got the Drobo 5N to replace my Drobo FS earlier in the week. I had an initial problem where it decided it didn't like the drives out of the FS, but a firmware upgrade cured that.

Backup times have gone down from a regular 2 hours and 15 minutes on the FS to just 20 minutes on the 5N. Quite an improvement considering its using the same WD Green drives which aren't particularly quick.

The Raspberry Pi's just keep plodding along. There have been a few updates so I applied them this week to all 5 Pi's.

I am holding off on getting any more Pi's until the Parallella turns up and we can see how it performs, although Parallella is more expensive than a couple of Pi's put together.

I still have not found any decent multi-Pi cases. There was one interesting one where they used some wood and glued HDMI connectors to it. The Pi's stand on their side using the HDMI connector to hold it in place. He had a router underneath the wood and on a 2nd layer had a 7 port USB hub to provide power.

12 May 2013

12th of May

Not much happening on the farm this last week. The Intel-GPU machines were running all week, well until the weekend. I went away for the weekend and forgot to enable work fetch so they just finished off what they had and sat idle all weekend. A bit of a nuisance but not worth worrying about.

Its been a bit warm during the day so the GPUgrid crunchers have been off.

Asteroids news
After their recent validator issue they have gone back to using the previous version they had.

Asteroids have been selected as one of the projects for the BOINC Pentathlon from the 14th to the 19th of May. Link to news: http://asteroidsathome.net/boinc/forum_thread.php?id=129

Albert news
Their Raspberry Pi 1.04 app turned out to be the same speed as their previous version. They have now release a 1.05 app that appears to be substantially quicker. My Pi's normally complete Albert BRP4U (Single DM) work units in 47 hours. With the 1.05 app they look like doing them in 31-32 hours. I say "look like" because they haven't finished the 1st one yet.

All 5 of my Pi's are running Albert work at the moment. After reaching my Asteroids goal of 1 million credits I decided to switch them over to Albert to help with their testing.

Parallella news
They have Linux up and running on their first batch of boards. Here is a picture of one running:

The latest news can be found at: http://www.parallella.org/

05 May 2013

5th of May

Farm news
Asteroids@home had a problem with their validator. It decided all work uploaded was invalid. The project staff have restarted it and it seems to be behaving now. At one point I had 380+ work units marked as being "validation inconclusive" and then it sends them to another user.

Despite this glitch I just passed the 1 million credits milestone for them about 15 minutes ago. I will now allow a spread of projects rather than concentrating exclusively on Asteroids.

With the cooler weather I am allowing the Intel GPU machines to crunch 24/7 rather than just overnight. My GPUgrid crunchers are also running but only overnight.

Hardware-wise I am looking at replacing the Drobo FS with a Drobo 5N. It is basically an updated FS with the addition of an mSata connector to allow an mSata SSD to be attached, thus making them faster.

28 April 2013

28th of April

Another month almost gone. Time flies when you are having fun.

In farm news this fortnight, not much done last week as I was away and the farm was running in minimal mode. That usually means overnight and minimal cache settings, so no crunching during the day. My internet provider (TPG Internet) has download limits (they have a peak and off-peak download limit, however their off-peak is only 6 hours - 2:30am to 8:30am).

Anyway I am still concentrating on getting Asteroids to 1 million credits for the moment. The last time I looked I was on 900,000. That would suggest maybe another fortnight (depends on how many machines and how long they are allowed to crunch) before I get there.

BOINC testing
We got 7.0.64 a bit over a week ago, there was a quick fix (7.0.65) for Linux and Mac's. This became the release version. I still need to update a few machines to it. 7.0.65 hasn't made it into the Debian repository yet (it still has 7.0.27) so the Raspberry Pi's haven't been updated either.

Nvidia GTX780 and 770 rumours
There are rumours doing the rounds that Nvidia will release a GTX780 and a GTX770 next month. The specs are still a bit unknown but the GTX770 sounds like it would make a good GPU cruncher. Having said that those people who rushed out to buy the Titan still don't have many projects to choose from. The GPUgrid apps fail on them.

Parallella update
They issued update #33 last week. They have some production boards coming in now so once they validate them it sounds like they'll be able to get into full production soon. It will be interesting to see how they perform using their 16 (or 64) core Epiphany chip as a coprocessor running OpenCL apps.

10 April 2013

10th of April

I haven't updated my stats in a while. At the moment most of the farm is concentrating on Asteroids@home. I am currently on 600,000 credits and I am trying to get to 1 million credits.

BOINC testing
We got 7.0.60 late last week. The main change in this is to correct an XML parsing bug that appeared with CPDN. They run a really old version of BOINC on their servers.

I expect there will some other minor changes before its ready for release. There was one particular crashing bug one of the guys found. I also raised a query as to why the Raspberry Pi doesn't report its CPU information.

The resource share bug still seems to be in this, however the client seems to alternate between projects so it may be doing the right thing over the longer term. In the 6.x versions it would run the appropriate resource shares of projects all at the same time. The 7.x version is designed to be less monotonous.

Project news - Einstein
I asked last week what is going to happen with the BRP4 single DM app that I am testing with the Raspberry Pi's. It looks like their current plan is:
  1. Introduce a BRP5 search using data from the Parkes Perseus Arm survey
  2. The current BRP4 search will become CPU only
  3. Allow the BRP4 search to do single DM work units

Project news - Seti
Their server relocation went surprisingly well. Apart from having staff available 24 hours a day to kick machines when needed they have better cooling and power. In addition they seem to have much better network bandwidth.

Looking at the new Cricket graphs it seems its running at 350Mbit instead of the capped 100Mb they had before. This means transfers are now much quicker and no more getting backed off. With this improvement I will now resume Seti crunching, especially once Asteroids gets to 1 million.

02 April 2013

2nd of April

Well its just past midnight as I write this, so we're into the 2nd of April.

BOINC testing
We got 7.0.59 at the end of last week. It has a fix to the resource shares issue that I indicated in previous posts. While things have improved it still has issues. There was a mention by Dr Anderson that there would be another fix required to make it behave.

I made donations to both Asteroids and GPUgrid, even though I haven't been able to run much GPUgrid work in the last 6 months. I hope this will enable both projects to continue their work. Unfortunately I can't claim tax deductions for either project.

Farm news
No much happening on the farm. I have been running 7.0.59 as mentioned above on a few machines to see how it behaves.

The Pi's continue on the way without a hiccup these days. I have attached all of them to both Albert and Asteroids as both projects support the Raspberry Pi. Albert is the Einstein beta test project, so usually as they get things tested they then put them across to the main project. I am not sure how long they will want to beta-test the Raspberry Pi application.

Seti server relocation
Just a reminder that the Seti server relocation will start in the next couple of hours and take at least 3 days. Personally I think it will take them a week to iron out any issues but we will just have to wait and see.

We don't know if the Cricket graph (it shows the network bandwidth usage) will be available once the move is complete. Its very useful to see how the network is holding up.

28 March 2013

Parallella design files shipped to manufacturer!

This is the Parallella board that we're all waiting for.

In their latest update they said:

"It took longer than expected, but the first revision of the Parallella board has been sent off to the manufacturer and we will have the first 10 Parallella boards built up and ready for testing by April 15th!"

We will see if they get them ready by April 15th. While I was a backer in the Kickstarter project I expect I am further down the order for the boards as they come off the production line - I didn't pledge that much. However it will be interesting to see what these things can do from a BOINC or number crunching point of view.

24 March 2013

24th of March

Just a quick update. The Seti and BOINC project computers will be moving to a co-location facility elsewhere on the Berkeley campus. That means both projects will be unavailable. They expect this to happen on Monday the 1st of April and to take 3 days to complete. Hopefully its not an April fools day joke.

Hopefully this will allow the project staff to concentrate more on the science and less on running machines. It may also give them access to a higher speed network connection in the future.

20 March 2013

20th of March

Not a great deal to report for the last fortnight. The weather is still warm but we had a couple of cool days where I had various machines running, at one point I had all the Intel-GPU machines plus the dual GTX660 machine running.

The Pi's are still crunching away for Albert and Asteroids.

Albert@home have started a test of a new FGRP OpenCL app. The current FGRP app is a CPU app so this is allowing GPU's to assist. It still spends a lot of time using the CPU so you'll need to dedicate a CPU core but as they tune the app this should improve. Its for ATI/AMD cards only at this point.

Asteroids@home have release an app for Android on ARM processors. They are the first project to officially support Android devices.

Seti network congestion
As previously reported their network connection is running at its 100 Mbit limit almost continuously. It seems there is a registry setting in Windows where you can enable a TCP option to improve performance on such connections. This option was suggested in May 1992. Unfortunately Windows doesn't have this option turned on by default but Linux and OSX do. For more information search for "RFC 1323".

To add the registry key cut the command below and paste it at a Windows command prompt. Make sure you do a "run as administrator". After doing this you'll need to reboot for it to take effect.

REG ADD "HKLM\SYSTEM\CurrentControlSet\services\Tcpip\Parameters" /v "Tcp1323Opts" /t REG_DWORD /d 3

BOINC testing
We got 7.0.56 this week. just some minor fixes but nothing noticeable from my side. This is a release candidate, although there have been a couple of fixes done since.

I still haven't convinced Dr Anderson that it doesn't honour resource shares despite providing some log files. It is doing this on 5 of my machines and seems fairly consistent. Its doing the exact opposite of the client emulator scenario that I ran.

There is an official port of BOINC 7.0.56 to run on Android. I don't have any Android devices so I haven't tried it.

03 March 2013

3rd of March

This weekend was windy and wet. Ideal weather for the crunchers really. I managed to fire up the GPUgrid crunchers for a few days as well as having the Intel-GPU machines going.
GPUgrid have some rather long work units in their long queue that have been keeping my GTX670's busy. The uploads haven't reduced in size yet, apparently they need to turn on a parameter on the server side the next time they create work units before it will take effect.
After a bit more experimenting with BoincTasks and the Raspberry Pi I have revised the instructions a bit. It seems you don't need to provide a password (well at least with BOINC 7.0.27 which is the version in the repo).
My faulty Pi was returned to element 14 who have issued a credit. Now all I need to do is buy another one.
Remember them, they have been making progress on their boards. They are similar to the Raspberry Pi except they contain a dual-core ARM processor and the Epiphany chip. They have 16 and 64 core versions of the Epiphany chip which is OpenCL capable. They are targetting a price of $100 for the 16 core version. Last week they released their documentation and a couple of weeks ago they released the source code for their OpenCL driver. The boards themselves have been finalised and now they are getting them produced in quantity
Setting up BoincTasks talking to a Raspberry Pi
  1. On the Pi you'll need to put the IP address and host name of the BT machine into /etc/hosts
  2. On the Pi you'll need to put the host name of the BT machine into /var/lib/boinc-client/remote_hosts.cfg
  3. Restart the Pi to pick up the above changes
  4. In BT (on your PC) you need to add the Pi. Click on the Computers tab and then on the Menu Bar Computers -> Add Computer. You need to use the IP address (not the host name) of the Pi and leave the password blank.

Points to note:
  • The Pi is invisible to the windows PC's on the network. The router will usually be able to see all of them but the windows machines only show other windows machines. The Pi's don't seem to have visibility of anything else on the network.
  • The Pi needs to have the IP address and host name of the BT machine in the hosts file. This is not good practice as IP addresses can and do frequently change. You may want to use static IP addressing

And here we have the makings of our own (rather slow) super computer running BOINC. Each node (Raspberry Pi) can be controlled via BoincTasks.

24 February 2013

24th of February

A flurry of activity on Saturday on the farm. The temp dropped a bit so I fired up the two GPUgrid crunchers and had them going until lunch time Sunday when it got too hot again. I left a couple of the Intel GPU machines running all weekend so they are making a bit of a contribution to Asteroids and Einstein.

BOINC testing
On the 13th of February we got 7.0.52. Nothing major, just some tweaks around GPU starvation when they are excluded from projects.

I am seeing some strangeness where it seems to favor the project with the lower resource share rather than the one with the highest share. That means I will have to run some logs and post them to the mailing list and also setup a simulation for the developers.

Pi news
The Pi's just get left running, I don't spend much time checking on them.

In my post of the 21st of February I explained how to setup BoincTasks so that it can talk to the Pi. I also got them talking to my proxy server. See details below.

Daniel Carrion posted a comment on my 21st of February post regarding SubsetSum where he has apps available for them via his blog.

Proxy server and Pi
The process went like this:

1. Make sure /etc/hosts has an entry for the proxy server. You'll need the IP address and hostname.

2. Add the following lines to the file /etc/environment
export http_proxy="http://proxyhost:port"

3. Create a new file in /etc/apt/apt.conf.d/ and name it "10proxy"; put inside
Acquire::http::Proxy "http://proxyhost:port"

Where proxyhost is the name of your proxy server and port is the port number its using.

21 February 2013

21st of February

Well last week was a bit cooler and I managed to have all the Intel-GPU machines running for about 3 days straight. This week is back to being hot so they are off again.

The Pi's have been running regardless of the weather. In fact the two new ones turned up on Monday this week. I got one of them going on the Monday and the other on Tuesday. At present 3 of the Pi's are running Albert@home and the other 2 are running Asteroids@home tasks.

From the screen shot you can see where I have added one of the Pi's to BoincTasks. I haven't updated the others yet

Setting up BoincTasks talking to a Raspberry Pi
BoincTasks is designed to monitor a number of machines. Unfortunately its a Windows app and so getting it talking to the Raspberry Pi's is not the simplest of things to do. I assume that you've managed to get BOINC running on the Pi and BoincTasks (BT) is also running on your Windows PC.

  1. On the Pi you'll need to put the IP address and host name of the BT machine into /etc/hosts
  2. On the Pi you'll need to put a password into /var/lib/boinc-client/gui_rpc_auth.cfg
  3. On the Pi you'll need to put the host name of the BT machine into /var/lib/boinc-client/remote_hosts.cfg
  4. Restart the Pi to pick up the above changes
  5. In BT (on your PC) you need to add the Pi. Click on the Computers tab and then on the Menu Bar Computers -> Add Computer. You need to use the IP address (not the host name) of the Pi and the password from point 2.

Points to note:
  • If you installed the boinc-manager on the Pi it will no longer be able to connect to the boinc-client. This is because it doesn't know the password. You can execute it manually and supply the password from a command prompt if you still want to use it, but you now have BT working so why would you need it?
  • The Pi is invisible to the windows PC's on the network. The router will usually be able to see all of them but the windows machines only show other windows machines. The Pi's don't seem to have visibility of anything else on the network.
  • Due to the 2nd point the Pi needs to have the IP address and host name of the BT machine in the hosts file. This is not good practice as IP addresses can and do frequently change.

I will see if the guys in the Raspberry Pi forums know how to work around these networking issues.

09 February 2013

9th of February

And here are the Raspberry Pi's. Not the best way to put them but until I organise a suitable case this is how they are running. I have two of them running Albert@home work and the 3rd is running Asteroids work at the moment. Once the apps have moved from beta test (on Albert) to the main project (Einstein) then they will transition over to there. The Albert work units are taking 48 hours to complete.

Daniel Carrion left a comment on last weeks blog regarding power supplies. Sure Electronics make 5 volt power supplies in 3A, 10A and 20A capacities. You can buy them on eBay. This would suit a setup of multiple Pi's (they call them a Bramble) where you don't want lots of power packs. There is some wiring required.


I have returned the revision 1 B model as faulty. It cost $10 just to post it back, probably would have been cheaper just to throw it away. While I was at it I have ordered another two Pi's.

Meanwhile in other news the farm has been off most of the week again. They get a quick burst of activity when its cooler and then off again.

BOINC testing
We got 7.0.48 this week. Just a few minor tweaks around the installer. It has issues with PAE (or service installs) under Windows, if your run it that way you'd better wait for another release.

There is also a 7.0.50 for the Android out. This is very much a work in progress. There are some issues about battery overheating they need to address, and of course the devices are usually even slower than the raspberry Pi.

02 February 2013

2nd of February

So after a warm week yesterday cooled off and we had a howling wind. Great for the farm though. I fired up the non-GPU cluster of machines and the GPUgrid crunchers.

Seeing as the GPUgrid crunchers have been off for a while they immediately went into high priority mode thinking that they wouldn't be able to complete the work in time. Now they are 75% through they have decided they don't need high priority. Hopefully I can get a few work units out before the weather warms up again.

BOINC news
We got up to 7.0.47 this week. Nothing major in this, mostly cosmetic changes and some tweaks to messages, etc. I have it installed on two machines, one non-GPU and one with a GPU. Actually the term non-GPU is not correct, they all have Intel GPU's integrated on the CPU chip.

Pi news
The 1st Raspberry Pi that I got is still having issues so I have put in a request to replace it with the supplier (Element 14) in Sydney. In the mean time I have ordered two more Raspberry Pies. I have some vague idea that I would like to make a small cluster of them and use Beowulf or HTCondor if I can find a version that will run on them. I know I can run BOINC on each one of them but what would be nice if I could get BOINC to submit work requests to the cluster.

I have also been looking around for cases and power supplies. There doesn't seem to be anything out there for multiple Pi machines.

While searching I did come across one setup where they are emulating a MicroVAX 3900 on the Raspberry Pi and its running OpenVMS. They had two of them clustered together. The link is:

My one working Pi is currently trying out Albert@home work. This is the beta test project for Einstein@home. They are testing smaller size work units just for the Pi and an optimised science application. Its currently about 75% through its first work unit and looks like its going to be about 3 days per work unit. This is on a "medium" overclock as set by Raspi-config.

27 January 2013

27th of January

Again hot is the word this week. Mostly in the 30's. As a result the majority of the farm was off although the non-GPU machines got a few days of work done.

Pi stuff
The Pi stayed running throughout this. I have ordered some heatsinks for the Pi. There are 3 components that get hot as you can see from the thermal image below

The USB ports are on the right side and the chip to the left of that is the network/USB controller. As you can see its the hottest, represented by the red colour. The square chip in the center of the board is the CPU. The one on the bottom is a voltage regulator or capacitor.

My 2nd Pi (revision 2, B model) has completed 4 Asteroids@home work units so far. Its averaging 67 hours at the moment. I have its overclock set to "medium" according to Raspi-config. Once the heatsinks arrive and the weather cools off I will look at increasing the overclock.

In case you missed it Daniel Carrion has a blog for the Raspberry Pi. On there he has links to a number of projects science apps compiled for the Pi with the necessary app_info and instructions on how to install them. The link is:
Thank you Daniel for doing these apps.

Suprisingly nobody has a case with fan forced airflow for the Pi. Of the designs I have seen there are aluminium ones, wooden ones, cardboard ones and plastic ones, but none with a fan. They all assume it doesn't get that hot, but if you put your finger on the network/USB chip it will burn you. I might need to come up with a case of my own, ideally holding a number of Raspberry Pis. The nearest one I have seen with multiple Pis is the Lego Supercomputer:

BOINC testing
7.0.45 came out yesterday. There are some tweaks around the work-fetch when excluding GPU's. I will put it on one machine and see how it goes.

Project news - Einstein
They have released a new BRP4 science app that can now use compressed data files. If you are running with an app_info you'll have to go and update it yourself.  The new app is able to handle both compressed and un-compressed data files They don't expect to turn the compressed files on until next week, however once they do it should reduce the downloads significantly.

19 January 2013

19th of January

We had a heatwave this week in Sydney. The temperature made it up to 45 degrees C. I had most of the farm off but left the Raspberry Pi running seeing as it doesn't generate much heat and takes a long time to process work.

Debugging Pi
I also spent a bit of time with the Raspberry Pi trying to work out why the network connection seems to disappear. I think it was caused by a combination of things. I tried different network cables. Nope not that. I tried a different (non-green) switch. No not that either.

I noticed it was using an IPv6 address and while my router supports it all the other machines have IPv6 disabled. To test it I plugged the old router in as it only does IPv4 and suddenly the network seems more stable but still drops out from time to time. I unplugged the powered USB hub and plugged the keyboard and mouse in directly and that seemed to fix that.

Seeing as I couldn't stay running the old router I then went looking for a way to disable IPv6 on the Pi. There is a file /etc/protocols that one can edit, so I commented out anything that mentioned IPv6. Then I plugged the current router back in to test it. All good.

To find out what address your Pi is using at a command prompt type: ip -d addr
To quickly test the network connection I just did: ping google.com

Just to make sure I also put the SD card into the older revision 1 model B and it too stayed working, so it looks like there is nothing wrong with it after all.

I still need to work out why IPv6 doesn't work on my network seeing as the router supposedly supports it, but it isn't urgent as my proxy server software is rather old and doesn't do IPv6 anyway - something else I need to look at updating.

13 January 2013

2013 reorg part 1

Spent the weekend reorganising things for the non-GPU group of machines. This is the outcome

On the left shelving unit the bottom beige box is a UPS. The white machine above it is the file server, with the Drobo FS on top. The other two shelving units have the non-GPU machines. Their specs:

  • Asus P8Z77-V LE motherboard
  • Intel Core i7-3770 @3.4Ghz
  • 8Gb of Kingston DDR3 memory @ 1600Mhz
  • 320Gb HDD (out of old Core2Quad)
  • DVD-ROM (also out of Core2Quad)
  • Seasonic 500 watt power supply (from Core2Quad)
  • Fractal designs Arc Midi case
As you can see we used what we could from the old Core2Quad's that these machines replaced. You can also see the KVM and monitor.

They released their app for the Raspberry Pi. The work units take an estimated 76 hours at stock speed (700Mhz). I have one crunching away at the moment and hopefully it will be done by the end of this week. I also did a modest overclock to 800Mhz.

Einstein@home passes 1 petaflop
Sometime shortly after January 1st 2013, Einstein@Home passed the 1 Petaflop computing-power barrier. To put this in context, according to the current Top-500 computing list (dated November 2012) there are only 23 computers that deliver this much computing power.

One Petaflop is 1,000,000,000,000,000 floating point operations per second.

11 January 2013

11th of January 2013

My first post for 2013.

Farm news
Not much happening on the farm due to the weather. Well thats not true, there isn't much crunching happening. There is however some reorganisation going on.

We had a really hot day during this week where the temperature got up to 42 degrees. That caused a lot of bush fires, some of which are still burning. More hot weather is forecast for next week.

I picked up a 2nd hand KVM just before the new year. I got a bunch of network cables as I didn't have enough long ones and all the short ones were in use. Some are colored so I can tell which cables are running the switches. I also got some more of the chrome wire shelving units to put machines on.

The idea is to have the GPU-less machines over on one side of the room on shelves with their own KVM. The GPU ones will be on the opposite side of the room (they already have shelves) along with their own KVM. They take up more space because the cases are much larger, being designed for high airflow.

This reorg has forced me to get rid of a lot of old boxes and useless clutter to make room. I am still doing this, having filled up the garbage bin for a few weeks running and still have more stuff to dispose of. There is a nice pile sitting in the middle of the lounge room floor.

Remember the KFSN4 motherboard that I couldn't get to work? Well I gave it to one of the guys at work and he has managed to get it going. He thinks I had the memory in the wrong slots. I had it working at one point but when I went to put it back into its case it refused to work.

Raspberry Pi
The 1st Pi that I ordered finally turned up. I only ordered it in August 2012. Its a B model, revision 2 and seems to work. It also seems to boot up a bit quicker than the old one, probably because it doesn't have to swap things out of memory.

I will now try and get the broken one replaced under warranty. Fortunately it was purchased in Australia so it has a 12 month warranty.

I still haven't got any BOINC apps on it yet as I was trying to work out why the 256Mb one wasn't working. I thought it might have been the SD card image or that I had stuffed up the Raspbian OS. I have proven that wrong by running the newer one using the SD card from the old Pi.