BOINC Pentathlon 2019

[leadeater]

@Windows7ge

Setting my big server to 98% CPUs dropped NFS task time by 1 hour, that was a good tip.

[GOTSpectrum]

Sorry about being MIA these last 10-12 hours or so. 

 

Thinks seem to be going well, Depending on how we do for the sprint we could see a nice finishing position this year! 

[leadeater]

8 minutes ago, Ben Quigley said:

Sorry about being MIA these last 10-12 hours or so. 

 

Thinks seem to be going well, Depending on how we do for the sprint we could see a nice finishing position this year! 

I really need to get better at multi instance bunkering, keep running out of tasks and server goes idle. Not a problem atm but I really did a terrible job previously.

[Egg-Roll]

8 hours ago, leadeater said:

I have 3 of these  I'll see how I can help you out soon. I don't believe anything specially is actually required, just need make sure all the correct power cables are connected and right ram slots and it 'should' power on.

You're right...

*my ears my poor poor ears*  I won't be leaving it on today as I'm running out of time, I don't have ssh access and when I do ip addr show it shows up as 127.0.0.1 maybe I put the eithernet cable into the wrong one, beyond that everything seems to be functioning perfectly fine. Once I get that fixed it's 100% ready

[Windows7ge]

4 hours ago, leadeater said:

@Windows7ge

Setting my big server to 98% CPUs dropped NFS task time by 1 hour, that was a good tip.

I noticed that the BOINC program doesn't really intelligently manage the CPU usage around important system processes so if you set it to 100% it will really try to use 100% leaving no space for idle processes or even what sustains the BOINC application itself and that kills your WU times.

 

I noticed that what % you can set it to varies based on the type of work as well. OpenZIKA I can only set it to 85% before I hit very high 90%'s. However SixTrack I can set to 90% before the overall system utilization hits high 90%'s so it's important to note you'll have to tune this based on the jobs for best output/performance.

 

I would expect 2% for the system probably isn't enough (but your OS install is probably more optimized than mine) I set it to such a way where I can see the ripple of not totally 100% being used and it bouncing around 98, 99, 100. Just that bit of free CPU time helps the WU's. Got my jobs from about 3~4hrs down to 2.5~2.75 hrs.

 

Linux seems to do this much better either than or single channel memory is hurting me this badly compared to quad.

[porina]

It probably comes down to "depends on the work" but I've never had to leave capacity for system. Work is supposed to run at low or idle priority so it doesn't interrupt other activity going on in the system. I guess it is possible that similarly low priority task would be held up somewhat, but as they're low priority they shouldn't impact immediate performance.

 

Maybe it could be ram bandwidth. I think that is an overlooked area myself and modern CPUs have ever more CPU potential relative to ram bandwidth, not a trend I like.

[Windows7ge]

16 minutes ago, porina said:

It probably comes down to "depends on the work" but I've never had to leave capacity for system. Work is supposed to run at low or idle priority so it doesn't interrupt other activity going on in the system. I guess it is possible that similarly low priority task would be held up somewhat, but as they're low priority they shouldn't impact immediate performance.

 

Maybe it could be ram bandwidth. I think that is an overlooked area myself and modern CPUs have ever more CPU potential relative to ram bandwidth, not a trend I like.

If I leave it at 100% the system gets a little laggy and my WU times suffer. If you don't have these issues that's great.

[porina]

4 minutes ago, Windows7ge said:

If I leave it at 100% the system gets a little laggy and my WU times suffer. If you don't have these issues that's great.

On daily use systems I do drop the CPU % for the better general responsiveness, but I haven't found an impact to pure crunchers. Generally for crunching, it is the overall throughput that is key, so that needs working out. For example, generally with HT/SMT compared to without, unit times go up quite a bit, but it is offset from doing more tasks in parallel.

[Windows7ge]

4 hours ago, leadeater said:

I really need to get better at multi instance bunkering, keep running out of tasks and server goes idle. Not a problem atm but I really did a terrible job previously.

This is why I brought up virtualization. It may hurt the job times a bit but it would keep you from running out of work at which point you might as well take the VM penalty.

 

I'm almost done bunkering SixTrack jobs but I've run out of WCG jobs to hold me over until the event starts so my solution. Spin up a VM or two to get more work and use the CPU what-so-ever until LHC@home starts then switch back over to full-native after the WU dump.

[GOTSpectrum]

If we can continue to manage 1-1.1m points in javelin for the next couple times around we should be able to overtake OCN... Keep it up guys. So far just keep doing what you're doing. 

 

I hope you are all bunkered up ready for the sprint though, I won't be taking part in the sprint myself down to trying not to stress myself out and we will still need some marathon production anyway.

[Windows7ge]

9 minutes ago, Ben Quigley said:

If we can continue to manage 1-1.1m points in javelin for the next couple times around we should be able to overtake OCN... Keep it up guys. So far just keep doing what you're doing. 

 

I hope you are all bunkered up ready for the sprint though, I won't be taking part in the sprint myself down to trying not to stress myself out and we will still need some marathon production anyway.

I've got Sprint jobs bunkered, Marathon jobs bunkered, and systems running on marathon in the meantime.

[GOTSpectrum]

7 minutes ago, Windows7ge said:

I've got Sprint jobs bunkered, Marathon jobs bunkered, and systems running on marathon in the meantime.

Good man!

[Windows7ge]

5 hours ago, Egg-Roll said:

You're right...

*my ears my poor poor ears*  I won't be leaving it on today as I'm running out of time, I don't have ssh access and when I do ip addr show it shows up as 127.0.0.1 maybe I put the eithernet cable into the wrong one, beyond that everything seems to be functioning perfectly fine. Once I get that fixed it's 100% ready

A quick look at the Intel .PDF for that series board doesn't say that either interface is dedicated IPMI so either should get the job done. What OS did you put on it?

[Egg-Roll]

17 minutes ago, Windows7ge said:

A quick look at the Intel .PDF for that series board doesn't say that either interface is dedicated IPMI so either should get the job done. What OS did you put on it?

Ubuntu because bsd refused to play nice with boinc. It was installed via my p4 but that shouldn't make anything change. It's the server version so no gui.

 

Maybe force Ubuntu to refresh network?

[Windows7ge]

1 minute ago, Egg-Roll said:

Ubuntu because bsd refused to play nice with boinc. It was installed via my p4 but that shouldn't make anything change. It's the server version so no gui.

 

Maybe force Ubuntu to refresh network?

What information does ifconfig give you? You may or may not need to install net-tools. It could just be a matter of needing to assign a static IP because 127.0.0.1 is your loopback/localhost address & DHCP may just be disabled.

[porina]

We get to turn off GPUs tonight? YES!

[Windows7ge]

2 minutes ago, porina said:

We get to turn off GPUs tonight? YES!

I really wish I could have kept mine going to contribute more here.

[porina]

1 minute ago, Windows7ge said:

I really wish I could have kept mine going to contribute more here.

If I had a way to manage heat I'd not mind either, but a few GPUs equals a whole bunch of CPU systems so I don't tend to run them as such often. It's been far too hot in my house the last few days and I'm really looking forward to getting temps down to a more tolerable level again.

[Windows7ge]

34 minutes ago, porina said:

If I had a way to manage heat I'd not mind either, but a few GPUs equals a whole bunch of CPU systems so I don't tend to run them as such often. It's been far too hot in my house the last few days and I'm really looking forward to getting temps down to a more tolerable level again.

My issue was just system stability. The computer kept hard resetting. If I went though the trouble of putting the cards back on air and rigged up a test-bench on an old motherboard I probably could have kept them going but them my desktop would have been unusable and getting it going again post-marathon would have been an all day project.

[GOTSpectrum]

1 hour ago, porina said:

We get to turn off GPUs tonight? YES!

Maybe for you, back to folding for my rigs

[leadeater]

5 hours ago, Windows7ge said:

I would expect 2% for the system probably isn't enough

2% of 192 threads works out to be 4 threads so it's enough for that system.

[leadeater]

4 hours ago, Ben Quigley said:

If we can continue to manage 1-1.1m points in javelin for the next couple times around we should be able to overtake OCN... Keep it up guys. So far just keep doing what you're doing. 

 

I hope you are all bunkered up ready for the sprint though, I won't be taking part in the sprint myself down to trying not to stress myself out and we will still need some marathon production anyway.

Opps I way over produced NFS, I still have a huge bunk on my large system I haven't submitted. It ran all last night not putting jobs through because I disabled network. I'll keep those for next throw 

[Windows7ge]

46 minutes ago, Ben Quigley said:

Maybe for you, back to folding for my rigs

After the Pentathlon is over and after all the VM testing I plan to do is done I want to leave a series of VMs running on both of my servers contributing to WCG. I plan to utilize the systems at either 50% load or up to 50°C whichever comes first. I like playing with this powerful equipment but most of the time the systems extreme resources sit unused so keeping in mind heat-output, power consumtion, my necessitates, & noise I'd like to keep them going just I can't have them running full-tilt.

 

18 minutes ago, leadeater said:

2% of 192 threads works out to be 4 threads so it's enough for that system.

Oh, you're right I forgot a percentage of a larger number is more. That translates to a smaller percentage.

[porina]

1 hour ago, Ben Quigley said:

Maybe for you, back to folding for my rigs

Maybe once I get the garden container sorted I'll do more all year around, but generally this is more a winter hobby where it doubles as wanted heating.

[porina]

Quote

15/05/2019 23:41:24 | LHC@home | This project is using an old URL.  When convenient, remove the project, then add https://lhcathome.cern.ch/lhcathome/

Tried to turn on LHC, wasn't expecting this...