17 April 2015

17th of April

Farm news
The weather has been cooler for most of this week so I have had the Intel GPU machines running climate models (still going they're up to 113 hours so far) and some Asteroids work. Asteroids have finally fixed their missing files issue so work is now flowing again.

CPDN announced that they will only target 1 particular platform (Windows, Linux or Mac) for each type of climate model in future to save on development and improve their reliability. I would think that it may be easier to issue the work units as VirtualBox VM images so they don't need to get involved in which operating system to target.


Intel driver update
Intel released driver 10.18.10.4176 for the HD4000 so I was trying it with Einstein. It actually seems to work. The last few releases from Intel haven't worked. I didn't do many work units but managed to get the BRP4 work units done and then some Parkes PMPS XT (aka BRP6) work units. The bad news is its quite a bit slower than the (recommended) 10.18.10.3621 driver. I didn't try it with Seti and have since gone back to the 3621 driver as its faster.


BOINC testing
We got an early look at the preference changes in 7.5.0. They seemed to work fine but I have suggested some cosmetic changes. Others have also asked for additional settings such as a "in use" and a "not in use" set of preferences. No word yet on them coming or not.


Windows updates
Got a few fixes again for patch Tuesday as its known. There was the usual run around and update the farm. Also a few for the Raspberry Pi's (Debian Jessie).

While that has been going on I have been trying to get the windows time software (w32time) to behave and keep the PC's clocks more accurate. Microsoft chose to do their own version of the ntp client that works somewhat differently from the standard ntp software. Anyway after fiddling with a few things and using google a lot I have them working as they should be.

03 April 2015

3rd of April

Farm news
The weather is cooler now so I am managing to get a bit of work done. Last week I managed to get a 3 of the Intel-GPU machines running CPDN work all weekend. They have some HadCM3S (short) work units that only take around 16 hours to process. The bad news is they produce 2 upload files each that are 64Mb, so then they kill the internet connection as they all try and upload at the same time. I restricted the number of uploads for them to try and improve things. This week I am running more of them, but only on two machines.

The 6 core 12 thread machines are running Seti work at the moment. I'm trying to get the Seti credits up to the Einstein credits (they are about 200,000 less). The only Einstein work I am running at the moment is on the Raspberry Pi2's and Parallella's.


Proxy server
I've been running Squid 2.7 for some years now without any updates, so the last month has seen me updating it to something more current and try and get it to behave. One advantage to this is I can finally use HTTP 1.1 (just as they release 2.0). Squid 2.7 only supported HTTP 1.0. There is some more fiddling and optimising I need to do that takes time. This was further complicated by getting a new router which has various firmware bugs and loses some settings when rebooted.


NTP
There was a security bug discovered with NTP (the Network Time Protocol) and the Linux guys were pretty quick with a fix. The problem was they now ignore user settings. That lead to some late nights and trying to work out why it seemed to ignore my config settings and do something totally different to what it used to do. The init script that starts it up uses its own dchp derived config file so I had to fix the ntp init script to point to my config file.

The fix version seems a bit behind the "official" release from the ntp.org website which is up to 4.2.8p1. I expect the Linux guys will update to that eventually.


Future purchases
I'm looking at the network infrastructure at the moment and what needs to be done to make it more secure and reliable. I suspect a new (dedicated) proxy server and possibly new file server may be on the shopping list. More on this when I have a better idea how to organise the network.

15 March 2015

15th of March

Farm news
We're crunching overnight at the moment. Not all of the machines as its still not cool enough overnight but some of them are getting work.

The Pi's and the Parallella's are still running constantly too.

This week saw a bunch of windows updates, further complicated by the fact most of the windows machines have been off for the last month, so a heap of updates to download and apply.

I installed the GTX970's into the GPUgrid crunchers and have run a "short" work unit on each machine (they take around 2 hours). The GTX670's will go up on eBay soon.


BOINC testing
We got 7.4.42 for Windows to play with. No major changes just some bug fixes.


Project news - Asteroids
They've been having issues with work units failing to download (server side issue) and the guy that runs it is working in another city so its been left alone. It also was getting a certificate error, but he has managed to fix that. Its out of work at the moment and we're waiting on more.


Project news - CPDN
They've been restricting their work units to different computers, so the ANZ climate models which I used to run are now restricted to Mac computers only. The EU climate models I can get on the Windows machines and they take around 9 hours. The remaining work units are restricted to Linux hosts.

01 March 2015

1st of March

Farm news
Everything is off except the Raspberry Pi's and Parallella's.. Today got up to 32 degrees, which is basically how hot the room with the computers gets. I really need to look for an alternative (air-conditioned) location for the computers.


Raspberry Pis and wisdom
Much like the last fortnight I have been concentrating on tuning the Pi2's to get the most out of them. That involved the fftw wisdom files which tell fftw what function choices are quicker. The Einstein app that runs on the Pi's and Parallella's will use it if its there.

I ordered some more copper heat sinks as I only had 2 sets, but 3 Pi2's. I also got some USB cables with power switches, as the only way to power them off is to unplug it. The power cables arrived two weeks ago and the extra heat sinks last week.

The timings I am getting for the Einstein BRP4 tasks are around 16.17 hours for Pi2 with wisdom and 13.83 hours with the wisdom and over clocking to 1Ghz. In contrast the B's and the B+ take around 31.5 hours. I have generated a wisdom file for the B+ as well (all the B's have been retired) and will have to wait for some results to see if it helps.

Generating a wisdom file can be tricky because the Einstein app has been compiled with fftw 3.3.2. Debian Wheezy has 3.3.2, Debian Jessie has 3.3.4 and Ubuntu Trusty has 3.3.3. The wisdom file has to match the version of fftw. For the B+ I had to put Wheezy on an SD card, boot up the Pi with Wheezy, generate the wisdom and then copy it over to Jessie.


Pi surgery
I also did a bit of surgery on one of the cases to see if it helps with cooling as you can see below.


The idea was the same as the older B model Pi that I mounted a fan on top, however thought I would try with just the grill and see how it goes. The Pi with this is running about 5 degrees (C) hotter than the ones without their top on. The case design is curved so its difficult to mount a fan.