I am doing some forum research... Can we get a hit at the overall "Stephan/xii5ku " hardware? It looks a bit crazy if it is not elsewhere used for commercial production of some kind!
I would just like a peak behind the curtain!
Before I discovered Distributed Computing as a hobby, I had two dual-processor computers with Xeon E5-2690 v4. I bought and built them myself and used them occasionally for simulations in my job. Such simulations were ran on an Ethernet cluster of office PCs before. Consolidation of this workload onto the SMP machines made these simulations quite a bit faster.
I posted photos of one of these computers
before this forum gained the feature of attachments.
When I got into DC, I was quite interested in
comparing performances of a variety of CPUs. But as I participated in one DC contest after another, I got tired of controlling computers with different performance profiles. And so I have lately preferred to build monocultures of several computers of the same type. This has helped me to retain some last shreds of sanity during contests like Pentathlon or Birthday Challenge.
Another thing which changed but already during the time when I started to get into DC, was that I am no longer using my own computers for my job now. (I am an employee, not a business owner.)
At this time I have several dual-processor computers for CPU based DC projects, and some dual-GPU single-processor computers (one triple-GPU computer) for GPU-based DC projects.
The exact total core count is: The same as many of us here have; that is, not as many as we would like to have, but more than a sensible and responsible person would have at home.
Anyway: One key step in my preparations to Birthday Challenge was to have a somewhat homogeneous pool of hardware, rather than a harder to control historically grown variety.
The next step was to figure out which of the WCG subprojects gave best PPD on my computers. Sounds like whoring for credits, but OTOH this also ensures that the WCG project gets the most use out of my computers, since the most fitting application is selected to run on this hardware. Sometimes I would run more or less meticulous measurements to find out PPD of different subprojects or different settings. But in preparation of this challenge I relied on prior knowledge and some simple crosschecks. This approach proved to be completely sufficient to select the optimum application.
Then I had to find out how many boinc client instances were required on each of the computers in order to be able to bunker for 10 days. And I had to "warm up" these client instances by completing an amount of WCG work in each of them, in order to get them recognized as "reliable" hosts by the WCG servers, which is necessary to get a reasonable number of tasks in progress assigned to them later. I combined the steps of determining the required number of client instances and warming them up, and in addition this warm-up gave a little contribution to the TeAm's involvement in the Thor Pure week.
Then a 10 days long routine began of loading some client instances with tasks, completing the work on them, loading more instances as late as possible, but not too late before the former client instances are done, start these next instances when the former are done, and so on ad nauseam.
I loaded new instances
as late as possible because I also wanted to be able to
unload them as late as possible. I figured that my TeAm mates who bunkered in front of the challenge too would unload the bunkers rather early in the challenge, making for a sharply decelerating production. I have a manic desire for acceleration, especially near the end of a contest. Therefore I loaded late and was able to unload late.
About the loading of a set of bunkers: Putting the respective client instances into the correct state, requesting work, and suspending network transfers as soon as all file downloads were finished, is a process which I automated to a good degree by means of some scripts. I nevertheless performed the loading of bunkers only when I was at home and awake, such that I could intervene if something unforeseen happens, like the internet link being down. This much restricted at which times of day I would perform downloads. Therefore it required some basic planning in how many client instances to fill when.
Filling a set of instances with work took about 50 minutes each time. Therefore, and because I usually had better things to do than to watch the work slowly trickling in, and because I had to use times of days for this when I was not always capable of concentrating on such mechanical procedures, it was crucial that I had automated this mostly, such that it was hard to make mistakes.
Whenever I had loaded more instances, I noted them on a list together with the dates and times at which the results of such a set of bunkers were due to be reported.
Another mechanical step which had to be repeated many times during the multi-day bunkering routine was to start computation in one client instance as soon as a former client instance on the same computer was about to finish all of its queued work. This step is automated by scripts here too. This procedure works reliably enough that it can run unattended for many days.
Finally, the uploading and reporting of results — in waves of each set of client instances which had the same reporting deadline — was supported by some simple control scripts too. I was present to watch this process for the initial two waves, to be sure that it works right. But subsequently I would let this happen unattended, in order to stretch these bunker releases over several hours during which I was asleep or was not at home.
I stretched each of the Tuesday...Saturday uploads over multiple hours (rather than uploading an entire set of buffers in a sudden burst) in order to obfuscate that TeAm AnandTech had bunkers in reserve all the time through the challenge. [Of course, those who know my modus operandi from contests such as Formula Boinc would have seen that my daily output on Monday through Saturday remained in the same ballpark, might perhaps have seen
my announcement of 10 days upfront bunkering, have seen Tony's estimation that Sunday could be interesting, and would have put two and two together.] — I unnecessarily stretched the Sunday uploads over several hours too, for effect. ;
-) This went on until it was already very late night here, therefore it was good that I had a script doing this while I could watch a movie and later read a book to kill time until the challenge results were up.
BTW, neither was it the first time that I maintained this routine during the entire 17 days before and including a WCG Birthday Challenge. Nor am I the only one who is doing such things. There was once a guy on another team who bunkered for
many weeks in front of a SETI Wow!-Event contest, on one or two computers.
Summary: The result which I contributed to this contest was facilitated by 1.) planning, 2.) good kit, 3.) some degree of automation.
PS, another word on my use of server CPUs, and comparably modern ones at that: A starving DCer would normally stick to less costly desktop CPUs or older server CPUs. However, electricity rates are high here, and I have cooling constraints, and limited room for computers. Hence my desire for good perf/W and good density.