BOINC Pentathlon 2019

[GOTSpectrum]

Just now, Ben Quigley said:

My plan next year is to set up each system with a VM for the GPUs as the way BOINC manages this on its own is bullshit. 

Actually it would probably be easier to set up VMs are the CPU cores and have the host use the GPUs but still...

[leadeater]

15 minutes ago, Ben Quigley said:

@leadeater

 

Would you be able to move a machine over to yoyo also, we have a real change of gaining points there, but I feel some of the other teams will buncker towards the end and I'd rather we tried to get some buffer. 

Moved everything but my work PC to yoyo.

[Egg-Roll]

15 minutes ago, leadeater said:

Mine are all Ubuntu Server 16.04 LTS

Why are you forcing poor little squirrels to do your bidding?

 

I'd of likely gone with Fedora instead. For the original servers propose no auto setup scripts exist for 18.04 which I'm actually happy about

[GOTSpectrum]

4 minutes ago, leadeater said:

Moved everything but my work PC to yoyo.

Thanks! 

[porina]

22 minutes ago, Ben Quigley said:

SMT must help better utilise the silicon otherwise it wouldn't be so prevalent on HPC chips is my thinking... 

 

the performance gain must be greater than the power usage gain otherwise it would be disabled on most server grade chips...

This is an area I do want to look into myself as I'm not aware anyone else has done similar, from an enthusiast viewpoint. My current understanding is that if SMT is used then it will take some power separate from the execution part. When I say used, I mean actively in use, not just enabled. If you run code in a way that SMT is present and enabled but not used, it is like it isn't there at all. I do that for known cases where it applies. In my earlier example it is worst case for SMT. The work does not benefit from it, but you're still paying the additional power cost for using SMT. In a good case, maybe you do get more work done than the increase in power consumption. I want to quantify that for myself, somehow.

[leadeater]

10 minutes ago, Ben Quigley said:

Thanks! 

Only problem with yoyo is I only get Siever jobs because I have to turn off ECM, the beast server doesn't have enough ram to handle the amount of jobs boinc want to run on it so it dies. So job times are around 8 hours, ECM is like 3 hours. My other servers could do ECM fine but then the big server would try and run them so...

[Windows7ge]

15 minutes ago, leadeater said:

..... shhh don't make me look stupid twice in one day 

First I did something dumb, then Ben Guigley admitted to saying something dumb, now you did something dumb (and I was wondering if you had configured them to use max resources because it doesn't by default...), we're all doing something dumb during this event. The important thing is get organized now, do better next year.

 

1 hour ago, Ben Quigley said:

@Windows7ge any change you can swap a heavy hitter machine over to yoyo for the next two days?

Is yo-yo just CPU? If so I can do that. Give me a while to figure it out.

[leadeater]

1 minute ago, Windows7ge said:

I was wondering if you had configured them to use max resources because it doesn't by default...

You did more than wonder, you told me something was off and I didn't go looking properly. You told me 2 days ago >.<

[jiyeon]

5 hours ago, leadeater said:

So I just figured out I'm a complete idiot, all my linux server were still set to the default max CPU of 60%.

4 hours ago, RollinLower said:

so basically, you've been crunching at almost half power all this time?

I just noticed my default max was also 60%... What's the benefit in going full 100%?

[GOTSpectrum]

7 minutes ago, Windows7ge said:

Is yo-yo just CPU? If so I can do that. Give me a while to figure it out.

No need now with Mr lead switching a bunch of his systems.

[leadeater]

1 minute ago, seoz said:

I just noticed my default max was also 60%... What's the benefit in going full 100%?

40% higher project output, it's just a simple CPU utilization limit.

 

 

Arrow where limit was changed form 60% to 100%.

[Windows7ge]

10 minutes ago, leadeater said:

You did more than wonder, you told me something was off and I didn't go looking properly. You told me 2 days ago >.<

I did?...I really don't remember.

[leadeater]

8 minutes ago, Windows7ge said:

I did?...I really don't remember.

Yea you said the numbers were off because for the number of servers I had you were matching, beating me actually at the time (you just didn't know it yet).

 

I just assumed because I was on old arch and only 6 core CPUs it was about right, how wrong I was.

[Windows7ge]

4 minutes ago, leadeater said:

Yea you said the numbers were off because for the number of servers I had you were matching, beating me actually at the time (you just didn't know it yet).

 

I just assumed because I was on old arch and only 6 core CPUs it was about right, how wrong I was.

Well it'll be a while before your new daily average becomes viewable on the site. What are you at now?

[Ithanul]

27 minutes ago, leadeater said:

Only problem with yoyo is I only get Siever jobs because I have to turn off ECM, the beast server doesn't have enough ram to handle the amount of jobs boinc want to run on it so it dies. So job times are around 8 hours, ECM is like 3 hours. My other servers could do ECM fine but then the big server would try and run them so...

May have to dig and do specific settings in the app_config per server.

 

I don't mess with that config file much, so I would have to go digging to see how one can set specific machines to only do certain subprojects per machine.

[leadeater]

7 minutes ago, Windows7ge said:

Well it'll be a while before your new daily average becomes viewable on the site. What are you at now?

I won't know either, these jobs are ~8 hours.

[jiyeon]

Looks like allowing WCG to share my data worked to get my BAM profile synced. 12.7K credits generated so far, but that was when I was on 60% compute power. I've now raised to 75% as I still use my PC for other things so I'll see how that turns out.

 

[Windows7ge]

Just now, leadeater said:

I won't know either, these jobs are ~8 hours.

Jeez. I've got jobs for OpenZika that are taking 4~5 hours but that might have something to do with those cores keeping the Linux VM running.

[Windows7ge]

6 minutes ago, seoz said:

Looks like allowing WCG to share my data worked to get my BAM profile synced. 12.7K credits generated so far, but that was when I was on 60% compute power. I've now raised to 75% as I still use my PC for other things so I'll see how that turns out.

 

Assuming the way utilization works is linear just add 15% to your current daily output and that should be more or less where you end up.

[leadeater]

7 minutes ago, Windows7ge said:

Jeez. I've got jobs for OpenZika that are taking 4~5 hours but that might have something to do with those cores keeping the Linux VM running.

That seems long for Zika but I get the same thing on the quad socket system. I was trying to dual instance and run two projects at once and limit the cores each could use but the tasks keep moving threads and clash so run times wen't up a fair bit for the tasks getting effected by that. I'd have ones doing 1 hour 30 like usual across that and all my other systems then others taking 3 hours.

[Windows7ge]

21 minutes ago, leadeater said:

That seems long for Zika but I get the same thing on the quad socket system. I was trying to dual instance and run two projects at once and limit the cores each could use but the tasks keep moving threads and clash so run times wen't up a fair bit for the tasks getting effected by that. I'd have ones doing 1 hour 30 like usual across that and all my other systems then others taking 3 hours.

In the Windows server I have jobs that at taking 2hrs (pretty normal) then I have jobs that are taking 4hrs+

 

This issue started as soon as I started the VM so I think it's to blame.

Meanwhile inside the VM everything's just hunky-dory

Hour-forty-five to two across the board. I think this is still better than having the 2nd cpu do nothing.

[Ithanul]

38 minutes ago, Windows7ge said:

Jeez. I've got jobs for OpenZika that are taking 4~5 hours but that might have something to do with those cores keeping the Linux VM running.

Odd, I notice getting around ~40mins per OpenZika in Mint Mate 18.3 (updated kernel) with a 1900X.  Though, that is not being housed in a VM environment.

My 1950X under a W10 LTSB gets around ~2 to 3 hours per OpenZika.

 

I may try to get the E5-2670V3 involved.  Seems a slight load on it don't bother the Minecraft server on it (been trying to get a darn git repo to build, but the blasted thing keeps failing to build the dev 1.14 papermc jar.....grrrrrrr).

[leadeater]

7 minutes ago, Windows7ge said:

In the Windows server I have jobs that at taking 2hrs (pretty normal) then I have jobs that are taking 4hrs+

 

This issue started as soon as I started the VM so I think it's to blame.

Meanwhile inside the VM everything's just hunky-dory

 

Hour-forty-five to two across the board. I think this is still better than having the 2nd cpu do nothing.

Could be worse, just checked the last few Zika jobs I let run to the end. Everything else is yoyo, these were just close enough to the end to not suspend/abort them.

 

 

8-10 hours lol

 

Edit:

Also Windows.

[leadeater]

2 minutes ago, Ithanul said:

Odd, I notice getting around ~40mins per OpenZika in Mint Mate 18.3 (updated kernel) with a 1900X.  Though, that is not being housed in a VM environment.

My 1950X under a W10 LTSB gets around ~2 to 3 hours per OpenZika.

Kind of varies a bit.

 

Below system is Linux 2x E5-2667

 

Fastest valid jobs:

Longest valid jobs:

[Windows7ge]

2 minutes ago, Ithanul said:

Odd, I notice getting around ~40mins per OpenZika in Mint Mate 18.3 (updated kernel) with a 1900X.  Though, that is not being housed in a VM environment.

My 1950X under a W10 LTSB gets around ~2 to 3 hours per OpenZika.

Yeah, no I threw Ubuntu on an SSD and ran it native on my 32 thread server and it's averaging hour thirty, hour forty-five, the VM overhead on the other server is there and noticeable but I could probably help it by trying a different disto I'll try Mint Mate in the future.

 

6 minutes ago, leadeater said:

8-10 hours lol

Lot of good 192 threads does you when the same number of jobs can be done with 1/4 the threads working 4x as fast. My 32 thread server would be on par with that behemoth if each job of yours took 8 hours.