Anyone else finding WUs from 155.247.166.xxx seem slow?

Moderators: Site Moderators, PandeGroup

Anyone else finding WUs from 155.247.166.xxx seem slow?

Postby DrBB1 » Thu Jan 11, 2018 3:15 am

Not quite sure if my PC has developed a problem or if there is something about the WUs from Temple that's a bit off. [Since I haven't seen any other posts about this, I have been assuming it's me up until now.] Further details follow.

I believe the vast majority of all the WUs I have been folding on my desktop PC since November have been associated with server 155.247.166.119 or 155.247.166.120. So, I was affected like many others when the Temple servers and the Stanford stats server weren't speaking to each other late last year, and I believe when the problem was fixed I received major credit for many previously completed WUs.

I noticed, however, since that problem was fixed, that my PPD on that PC has gone down about 50%, from over 10,000 to a bit over 5,000. Initially I attributed the drop off to updating my OS from Windows 7 to Windows 10, but there has been little, if any drop off in the performance of software I run in the foreground. Also, when I have no foreground applications running, FahCore_a4.exe is using 92-98% of my CPU, a bit less than it did under Windows 7 (presumably because Windows 10 has more overhead). That said, it doesn't seem reasonable at first glance to believe that such a small drop off in CPU availability should result in such a large reduction in PPD. Indeed, since I upgraded to Windows 10, I have eliminated some apps that were using a bit of RAM and CPU, so F@H has less user-initiated software competition. I have checked the PC several times for malware and crapware, finding nothing after my initial purge of Microsoft bloat that loaded by default with Windows 10.

Because all the recent WUs I have received are from the two Temple servers, I can't determine if the drop off in PDD is a PC issue, something has changed regarding the base points allocated to WUs in general so points are lower across the board, or whether the only the WUs from Temple offer fewer points—or something else entirely. I am hoping someone has some inside info, or can point me to a Forum post I missed about this issue, (e.g., it's a known issue with Windows 10, or there is a a particular setting I should check in F@H or Windows 10. [I have confirmed that my slider is set to fold at 100% capacity]. Any other suggestions would also be appreciated. Thanks in advance.
========
DrBB1
DrBB1
 
Posts: 159
Joined: Wed Mar 26, 2008 12:30 am
Location: SE PA

Re: Anyone else finding WUs from 155.247.166.xxx seem slow?

Postby Joe_H » Thu Jan 11, 2018 3:24 am

Part of your drop is that most of the WU's currently coming from the two Temple servers use the A4 core. It is older and less efficient computationally than the newer A7 core. There are some projects using the A7 core, but they appear to have provided fewer WU's over the last month or so as compared to earlier in the Fall. As the A4 projects are finished up and newer ones started that use A7, the proportion should change and your PPD rise again.

You have noticed the reduced CPU usage associated with a change to Windows 10. Even though small, the reduction in throughput will have a non-linear effect on completion time and a resulting reduction in PPD. You may be able to reduce the OS overhead if you can identify services that MS starts by default that you don't actually need running on your PC.
Image

iMac 2.8 i7 12 GB smp8, Mac Pro 2.8 quad 12 GB smp6
MacBook Pro 2.9 i7 8 GB smp3
Joe_H
Site Admin
 
Posts: 4188
Joined: Tue Apr 21, 2009 4:41 pm
Location: W. MA

Re: Anyone else finding WUs from 155.247.166.xxx seem slow?

Postby bruce » Thu Jan 11, 2018 9:07 pm

To add to what Joe said, there's an issue of AVX support.

Older CPUs did not support the AVX instruction set while more modern ones do. Both older and newer CPUs supported SSE. Both SSE and AVX instructions allow multiple operations to be processed concurrently, significantly increasing the speed that the floating point operations can be processed but AVX allows for increased parallelism in those operations.

Bottom line: FAHCore_A7 is more efficient if you have a CPU that can use AVX. (For those with older CPUs, FAHCore_a7 can also use SSE.)

New projects will most likely be started using FAHCore_a7 but for reasons of scientific validity/repeatability, projects started using FahCore_A4 must be completed using FAHCore_a4, so those projects will continue until they are completed.
bruce
 
Posts: 21564
Joined: Thu Nov 29, 2007 10:13 pm
Location: So. Cal.

Re: Anyone else finding WUs from 155.247.166.xxx seem slow?

Postby Joe_H » Thu Jan 11, 2018 10:14 pm

To add one point of clarification, both Bruce and I have mentioned the word "projects" in relation to WU's running on either the A4 or A7 CPU core. That word has two folding related meanings that might not always be obvious. First it can mean a specific research project being run by a graduate student, faculty member, or other researcher. Within that project they may run multiple folding Projects to examine different aspects of the protein(s) being folded. That is a second meaning of the word as connected to folding.

So you may see a new folding Project come out, but it still uses the A4 core because it is from the same research project. A different researcher in the same lab might be starting up a new research project, and use the newer A7 core. Or they might end up using the A4 core if the results need to be compared and validated with results from another research project being run there on the A4 core.
Joe_H
Site Admin
 
Posts: 4188
Joined: Tue Apr 21, 2009 4:41 pm
Location: W. MA

Re: Anyone else finding WUs from 155.247.166.xxx seem slow?

Postby KWSN_Sir_Frank » Fri Jan 12, 2018 11:06 pm

An adjustment to Credit Granted would seem to be in order: normally my two Q-9500s crank out about 300 credits per hour each...
but produce only about 150-175 credits per hour on the workunits from Temple...
KWSN_Sir_Frank
 
Posts: 10
Joined: Sat Jan 23, 2016 1:03 pm

Re: Anyone else finding WUs from 155.247.166.xxx seem slow?

Postby bruce » Sat Jan 13, 2018 3:36 am

the speed (and the points) for WUs running with SSE have not changed, nor should they. FAH never adjusts points once they're earned ... and especially since the science you're producing hasn't changed.

Enabling some machines on some projects to produce science faster with AVX is an enhancement to FAH and when you do produce work faster, you do earn more points. You can also earn more points with a GPU because they complete more calculations in less time.
bruce
 
Posts: 21564
Joined: Thu Nov 29, 2007 10:13 pm
Location: So. Cal.

Re: Anyone else finding WUs from 155.247.166.xxx seem slow?

Postby KWSN_Sir_Frank » Sat Jan 13, 2018 5:50 am

OK - did not realize how much difference there was between the A4s and the A7s...i just received an A7 and it is zipping along at
a rate of 400 credits per hour...

My apologies for the lack of understanding...now it makes more sense...
KWSN_Sir_Frank
 
Posts: 10
Joined: Sat Jan 23, 2016 1:03 pm

Re: Anyone else finding WUs from 155.247.166.xxx seem slow?

Postby JimboPalmer » Sat Jan 13, 2018 8:20 am

Any research that already used a4, needs to stay on a4 so the results are consistent.
I think all truly new research is using a7.

a4 dates from about 2010, so it not only works on older CPUs, it uses more primitive algorithms on newer CPUs.
a7 drops some older (Pentium 3?) CPUs, but it does more science on many older CPUs. On 'new' CPUs that support AVX2, it uses newer instructions to do even more science.
AVX2 is on Intel, starting at Haswell, (2013) on AMD starting at Excavator models. (2015)
Tsar of all the Rushers
I tried to remain childlike, all I achieved was childish.
A friend to those who want no friends
JimboPalmer
 
Posts: 654
Joined: Mon Feb 16, 2009 4:12 am
Location: Greenwood MS USA


Return to Issues with a specific server

Who is online

Users browsing this forum: No registered users and 1 guest

cron