13th WCG Birthday Challenge

[StefanR5R]

OK, measuring PPD on my Linux hosts will provide value mostly to myself, as I have disproportionally many cores driven by Linux instead of Windows, and Linux versions of several WCG applications are said to perform different from (better than) the Windows versions.

WCG publish average run time and average WCG-points per hour-of-run-time themselves, averaged over all hosts from all contributors. Links found thanks to conversation with Tony.

MIP .......... 1h48m, 199 p/h
SCC .......... 1h38m, 180 p/h
Zika ......... 2h00m, 183 p/h
HSTB ..... 10h43m, 177 p/h
FAAH2 ... 12h49m, 181 p/h
OET ......... 1h29m, 179 p/h
MCM ....... 4h18m, 181 p/h
FAAH ....... 3h01m, 122 p/h​

So far I measured SCC, Zika, and OET on my Linux host and got results far above those averages. That's probably the Linux effect. I have read a few times that some WCG contributors with Windows hosts run Linux VMs just for WCG. Additionally, while my core clock isn't particularly high, it may still be somewhat above the average across all WCG contributors' cores. Edit: My core clock is probably somewhat below the average across all WCG contributors' cores.

Linux BDW-EP @ 3.2 GHz:

SCC .......... 1h07m, 298 p/h
Zika ......... 1h06m, 338 p/h
OET ......... 0h17m, 360 p/h​

PS: 7 WCG points = 1 BOINC credit a.k.a. cobblestone

 

[StefanR5R]

for my threadripper for mapping cancer markers, it looks like about 3.05 hours for 114 points. Sound right ?

Sounds right. Your run time is quicker than the global average (4.31 h), and if the 114 points are BOINC credits from the "My results" table, then that's 798 WCG points, therefore 262 WCG points per hour (better than the global average of 181 WCG points per hour).

 

[ao_ika_red]

Kudos to Stefan and others for your benchmark. I'll use the first week of November to bench my machine.
Hopefully, when WCG Challenge start, the rain will be more frequent than in recent weeks so I don't have to face another thermal problem.

 

[Ken g6]

FAAH might be low because I think it runs on Android ARM devices.

 

[StefanR5R]

@Ken g6, that makes sense, thanks.

Then the rather close points per hour of the other projects are certainly not coincidental. Users of the new MIP subproject complained in the WCG forum that their PPD gradually declined from an initially high level. I suspect this is the WCG project gradually adjusting MIP PPD to the level of the other subprojects.

 

[StefanR5R]

I just started Microbiome Immunity Project (MIP) on three Linux hosts and one Windows host. The Linux hosts returned their first few valid results. (This subproject has quorum=1 even on hosts which never did MIP before.) The Windows host had a large number of errors, all of them right at the start of the respective task, and all of them with a result log containing something like this:

failed to create shared mem segment: minirosetta Size: 25029688
Unhandled Exception Detected...
- Unhandled Exception Record -
Reason: Access Violation (0xc0000005) at address 0x01CE2F10 write attempt to address 0x017D7EC9

There are various problem reports posted in the MIP forum, but I haven't found reports of this one. The affected host has 146 errors vs. 1 valid and 1 in progress at the moment. So for me, this one is not worthwhile to run on Windows for the time being. YMMV. No errors on Linux, except for one transient download failure.

Other problems:

• Help Stop TB was out of tasks when I tried (on Monday).
  
• FightAIDS@Home was out of tasks too when I tried (today).

 

[crashtech]

I just started Microbiome Immunity Project (MIP) on three Linux hosts and one Windows host. The Linux hosts returned their first few valid results. (This subproject has quorum=1 even on hosts which never did MIP before.) The Windows host had a large number of errors, all of them right at the start of the respective task, and all of them with a result log containing something like this:

There are various problem reports posted in the MIP forum, but I haven't found reports of this one. The affected host has 146 errors vs. 1 valid and 1 in progress at the moment. So for me, this one is not worthwhile to run on Windows for the time being. YMMV. No errors on Linux, except for one transient download failure.

Other problems:

• Help Stop TB was out of tasks when I tried (on Monday).
  
• FightAIDS@Home was out of tasks too when I tried (today).

None of my MIP WUs are erroring or showing invalid, AFAICS; All run Windows 10. That specific Windows host of yours may have some kind of OS/driver problem.

 

[TennesseeTony]

I can't believe it's been so long since WCG had a GPU app.

Out of the 600 MIP tasks listed for me, only one shows as failed. It has been a few days since I've ran WCG though, perhaps you were 'lucky' and got a batch of bad workunits, Stefan.

 

[StefanR5R]

@crashtech, unless I just got a bad batch as Tony suggested, the application obviously has a problem with this particular host. But I also have no doubt that the host itself doesn't have a problem, as it runs countless other projects (including the other WCG subprojects) and stuff outside of DC just fine. Meanwhile I loaded 8 MIP tasks onto another Windows host (both are Win 7 Pro) and they are now in progress and likely to finish OK.

 

[StefanR5R]

Turns out MIP's problem wasn't particularly with my host, but with how I used it. The shared memory allocation error happens when I try to run MIP in more than one client instance at a time. If only one client instance runs MIP, it works. It may furthermore be dependent on the number of client instances active, or on the timing between starting jobs in different instances. (The problem does not exist on Linux.)

 

[StefanR5R]

Here are more measurements. I always ran just one application, and used all logical CPUs, hyperthreading enabled. (On Windows I set one processor thread aside for background programs.)

WCG points are 7 times of BOINC credits, and are given per processor thread used.
5-percentile = 5 % of the tasks went faster than this
95-percentile = 5 % of the tasks went slower than this
cv = coefficient of variation

MIP = Microbiome Immunity Project v7.11
SCC = Smash Childhood Cancer v7.08
Zika = OpenZika v7.08
OET = Outsmart Ebola Together v7.19
MCM = Mapping Cancer Markers v7.36

SCC, Zika, and OET are using the VINA code base (whatever this means) which is known to perform a lot better on Linux than on Windows.

First a comparison between subprojects on one and the same host:

CPU, clock, OS: E5-2690v4 @ 3.2 GHz, Linux

sub-      task duration,              WCG points/h,     sample
project   mean (5-, 95-percentile)    mean     cv       size
--------------------------------------------------------------
MIP       1.1 h  (0.6 h, 1.8 h)       200 p/h  (4 %)      120
SCC       1.1 h  (0.9 h, 1.3 h)       300 p/h  (8 %)       30
Zika      1.1 h  (0.9 h, 1.2 h)       340 p/h  (6 %)       30
OET       0.28 h (0.18 h, 0.40 h)     360 p/h  (6 %)       75
MCM       4.0 h  (3.4 h, 4.1 h)       210 p/h (11 %)       60

Remember that WCG's published points/h, averaged across all contributors, is at the order of 180 points/h, except MIP with 200 points/h (see post #26).

Now a comparison between different hosts and OSs, for the same subproject:

subproject: MIP (Microbiome Immunity Project v7.11)
                            task duration,              WCG points/h,     sample
CPU       clock   OS        mean (5-, 95-percentile)    mean     cv       size
--------------------------------------------------------------------------------
E5-2690v4 3.2 GHz Linux     1.1 h  (0.6 h, 1.8 h)       200 p/h  (4 %)      120
E5-2696v4 2.8 GHz Linux     1.2 h  (0.6 h, 2.1 h)       180 p/h  (7 %)      120
E5-2696v4 2.8 GHz Windows   1.3 h  (0.7 h, 2.0 h)       160 p/h  (3 %)      120

MIP uses the Rosetta code base (same as Rosetta@home), and performs somewhat better on Linux than on Windows too, but the difference is not as large as with the VINA applications.

 

[StefanR5R]

FAQ on limits of tasks in progress:
https://www.worldcommunitygrid.org/forums/wcg/viewthread_thread,28636
New clients get only 5 tasks per logical CPU, until they complete (and validate?) their first task.
Then the quota is raised to 80 tasks per logical CPU and 2560 per client. (The latter limit is ineffective because the client limits itself to 1000 runnable tasks.)

I have a 4C/8T client on which I set <ncpus>13</ncpus> during download. It got the message

This computer has reached a limit on tasks in progress​

already after it fetched 910 tasks. IOW WCGs limit for trusted hosts is 70 tasks per logical CPU, not 80. So I increased ncpus to 15 in order to receive the 1000+ tasks that the client itself accepts. I decreased it to 8 = the real thread count of the processor after I was done downloading.

(See the wiki for documentation on ncpus. IME it is harmless to run with too large ncpus for a while on Linux. But on Windows 7 it often causes the host to become unresponsive; even ctrl+alt+del could be delayed by a large lag.)

 

[TennesseeTony]

2 days and 20 hours until the fun begins!

If you haven't already, it's time to convert your running tasks to WCG, load up 3-4 days of tasks (or more, it's a 7 day race [minus one minute]), and then turn off network activity to build up the bunkers.

Let's hope xii5ku's 6 days of bunkering, then 7 days of bunkering after the start for a final bomb, is as amazing as his performance at Universe!

As for me, I actually bunkered more than I could chew, and aborted what I know I couldn't complete a few days back. So my hoped 1,000,000 bunker will fall short I think.

 

[ao_ika_red]

My whole fleet is ready. Don't expect much from 2 Steamroller modules and 2 Penryn cores, though.

 

[crashtech]

I'm experimenting with some VMs to see if a few tasks can be ready for the 16th.

 

[Markfw]

Watch your tasks while bunkering. All of mine are now running high priority which has a tighter deadline. If that happens, you have to make sure and report them before the deadline.

And less than 8 hours until the start !

 

[StefanR5R]

Watch your tasks while bunkering. All of mine are now running high priority which has a tighter deadline. If that happens, you have to make sure and report them before the deadline.

Definitely keep the deadline for reporting. Some projects give a bit leeway for late reports, but WCG generally doesn't, says their community-maintained FAQ.

The "running high priority" state is enabled by the boinc client following some complicated rules. I have seen it often even with work queues which the host was clearly able to finish way before the deadline. I think hosts with lots of downtimes in the past are doing this, but there are surely many more reasons. — Actually, the WCG FAQ has a long article on the subject of jobs running at high priority too.

And less than 8 hours until the start !

And then it's 7 days of relentless Birthday partying!

 

[ZipSpeed]

Finished my Ryzen 7 1700 rig just in time to be added into the race. I still have a 2600K and 3770K at home that I haven't spun up yet, but all the rigs in my sig should be going into the race. Plus I may be able to add some more cores at my work.

 

[TennesseeTony]

AWESOME!!!! Thanks ZipSpeed!

Just over three hours until the start.

@Markfw I hope you're feeling okay after yesterday's round of Chemotherapy, friend.

 

[TennesseeTony]

Last year's thread if you're bored: http://www.portvapes.co.uk/?id=Latest-exam-1Z0-876-Dumps&exid=threads/12th-wcg-birthday-challenge.2488905/

Some stats links to keep open or bookmark:

Standings: https://www.seti-germany.de/wcg/stats_team_all.php?proj=wcg&lang=en

TeAm Stats: https://www.seti-germany.de/wcg/stats_team.php?TeamIdWCG=SHNW8R84S1&proj=WCG&year=2017&lang=en

 

[Markfw]

AWESOME!!!! Thanks ZipSpeed!

Just over three hours until the start.

@Markfw I hope you're feeling okay after yesterday's round of Chemotherapy, friend.

I am feeling "ok" but barely functional. They gave me a 200% shot.

BTW, that stat link works for current, but do you have any record on how each team member did last year ? I think I just started on WCG last year, and I can't find historical stats like F@H.

 

[IEC]

It's 12:04AM UTC 16 Nov according to Google. Are we good to start uploading results?

 

[crashtech]

The counter is counting up now, so I think we are good!

https://www.timeanddate.com/countdo...16T00&msg=13th WCG Birthday Challenge (start)

 

[TennesseeTony]

Individual stats weren't provided last year Mark. About the best we can hope for this year is perhaps screen shots from Free-DC's team page. I can tell you that you SUCKED last year though. hee hee! And now you're making an incredible million a week at WCG, on top of the bazillion per week you make at Folding. (and thank you so much for it!) [both]


http://stats.free-dc.org/stats.php?page=team&proj=bwcg&team=16700

Let's see if this works: (IMG tags aren't working right on the site, you'll have to click)

Starting stats: https://ibb.co/nkOTdR

 

[TennesseeTony]

I dozed off watching "RoadKill" on youtube, so my uploads are still ongoing, but the TeAm started off in 6th.