| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Completed Research
Forum: Help Conquer Cancer
Thread: [RENAMED] Some concerns regarding the HCC project (page fault and poor performance, in particular (but not only) by multi-core hosts; #cores>2) |
| No member browsing this thread |
|
Thread Status: Locked Total posts in this thread: 210
|
|
| Author |
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
WHOA!!! Everyone hit the brakes... I'm interested that Sekerob mentioned that this can also be an L2 Cache problem. Which, if I'm right, is only resolved in the BIOS setup menu. Also, is everyone on the same page as far as running WCG on a separate partition, even if your machine(s) run WCG exclusuvely? It's that doggone kernel that keeps trying to override your operating systems' natural tendency to want to be #1. BOINC, BSD, UNIX and the like operated originally in large server environments, where neither Windows, nor MAC used to tread. I utilized these applications and languages on a Sun Solaris system years back, and the only hassles was when we used the VI interface incorrectly. That was it's text editor. I strongly believe that if one wishes to run all of the available projects, one must be willing to tweak their preferences, both in the BOINC location, as well as their own system environment, to agree. And the more projects one is willing to run, the less robust ones' statistics will be. I am of the strong opinion that HCC is a single core, exclusive run. That means standalone. Because of the way the it behaves as encoded. And I bet if it is run on a separate partition, their may be a turnaround in opinion. The opinion stated here is the opinion of the writer, solely intended to enliven the technical discourse regarding our crunching community. Let's talk shop people! It's about principles, not personalities. Let's be civil with one another.    Eh ?  |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Courine, if you have a problem with what I say, please don't ignore it then bring it up an entire year later!
I merely wondered, all that time ago, whether the site wasn't archived already. So I checked for you, and reported the results. That's what I do. |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Can we return our focus to the computational problem here. As I understand it, people working on HCC tasks have observed a large percentage of CPU time spent in the kernel. I typically see around 25% when looking at the CPU Usage History graph(s) in Windows Task Manager. I am running Windows XP SP2, and I have observed this with single core and dual core machines.
Probably related is the observation that Page Fault Counts are high, in excess of 10^9. So there is a potential to increase the number of HCC tasks completed per day by 25%, based on my observations. Others running quad core and two quad core machines are seeing even greater inefficiencies. Many of us here think this is a large enough problem to get some attention, and we are getting frustrated. Please, could we get one of the programmers to at least look into this problem and provide an assessment of whether the code can be improved with a reasonable effort? Steve |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
We have covered the technical aspects already. Numerous times.
The techs know about the problem, and are working on it. I will ask the techs for a status update in a while, if they don't give us the information soon. Steve, I appreciate your contribution, but you are covering old ground. We have already had a number of expert opinions, and seen various test results. I'm sorry to say the relevant posts are scattered through a number of threads, but if you are interested enough to contribute to the discussion, then finding them shouldn't be a problem. The techs took a first glance at the problem as soon as I reported it, before the project launch (on the 2nd of November). I don't think you are giving them enough credit. |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Hi steveleg,
I have been taking for granted that some of the debug time is going into HCC. This whole month is dedicated to debugging current projects. The programmers have not been talking about what bugs they are working on. I suspect that they think that if they mention the problems that concern them, then they will start getting many more complaints about each issue that they mention. Looking at a problem is not the same as solving it. Uplinger mentioned that they had already looked at the HCC page faults when the project was first released. Later last year they were working on the Linux version of HCC. I am positive that they have not suddenly abandoned HCC. Lawrence |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Didactylos, Lawrence,
Thank you for your reassurances. I'm pleased to hear that this problem is getting attention, and I look forward to getting a status communication and possibly a new, more effecient release of the HCC code. Steve |
||
|
|
teletran
Senior Cruncher Joined: Jul 27, 2005 Post Count: 378 Status: Offline |
Is it possible that these page faults/performance issues could lead to system freezes? I've tried running HCC work units in the recent past and on each occasion I've had my computer freeze (while concurrently running a video). There are no other changes to the system except that an HCC work unit has been running at the time. Might be coincidence but it happened each time I was running HCC. It may be totally unrelated but on the chance that it shows other system effects than just on the work unit itself, I thought it important to bring up here.
----------------------------------------I've already had to stop receiving HCC on this machine. For reference, I'm on an Athlon 64 dual core pc. |
||
|
|
knreed
Former World Community Grid Tech Joined: Nov 8, 2004 Post Count: 4504 Status: Offline Project Badges: 
|
Sorry for the silence. We are investigating this problem. We are currently running some pair-wise tests to determine what is causing the slow down. What I mean by pair-wise is that we manually sending a workunit for FightAIDS@Home and a workunit for Help Conquer Cancer to the same two computers that have the characteristics that we want to check. For example here is the outcome of one of these tests:
Computer #1 is my laptop. It is running a Intel Pentium M running at 2.0GHz with 2GB of DDR Ram Computer #2 is my home desktop. It is running a AMD Athlon 64 X2 5200+ running at 2.7GHz with 2GB of DDR2 Ram FightAIDS@Home workunit 'faah2961_ZINC00000480_xMut_md02740_00' had the following results: Computer #1 ran the workunit in 7.6 hours and claimed 83.238 credits for the workunit. Computer #2 ran the workunit in 6.1 hours and claimed 94.623 credits for the workunit (20% faster then the pentium M) Help Conquer Cancer workunit 'X0000046720001200502241630' had the following results: Computer #1 ran the workunit in 7.8 hours and claimed 85.014 credits for the workunit Computer #2 ran the workunit in 8.2 hours and claimed 128.109 credits for the workunit (5% slower then the pentium M) The next step we are going to do repeat this test but force the AMD dual core to only run one workunit at a time to see if eliminates the drop in performance for the dual core machine. |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
What might help some systems slightly, but nothing huge I would imagine, is to put the swap file on to a second physical drive if one is available. That way the swap file will have its own read/write heads separate from the rest of the system. This old machine has been like that since I built it. Probably not worth bothering about for most people though.
|
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
These are soft faults, not hard faults. We are pretty certain that virtual memory and drive configuration are irrelevant.
The configuration you suggest does improve overall system performance a little (I do it myself, where possible), but it has no bearing on HCC. |
||
|
|
|