Predictor latest news

Distributed Computing Projects Other DC Projects
1

11/28/04
A change of guard is underway: Michela, who has contributed immensely to the P@H project is now off to begin a new life on the CS faculty at UT El Paso, congratulations Michela! We have recruited David B. to takeover P@H upkeep and continued development of Predictor. David has more than 8 years experience in IT and is a great addition to the team. Welcome aboard David! Michela will continue her involvement with the aim of brining new projects relevant to biomedical research on-line with P@H.

It WILL be up again very soon. Don’t forget Chuck’s advice…be gentle. Don’t swamp them with every host you’ve got right away.

2

:xfinger: all goes well :smiley:

DT.

3

yep…The most benefiacial to the human rac I believe this project is…hope I can get a place some time soon…

4

Update

Not going to happen quite so soon after all I am afraid… :frowning: The changing of the guard is taking longer than expected. … have to be patient. :wall:

5

Followup.

Yes, Ike is right… our ‘Changing of the Guard’ is not going as quickly.
as expected. It is the end of semister and this brings everyone to
‘overload’ status. The new “Mr Predictor” is ready to assume duties,
but not been given the authorities and full training required.

Some good info I do hear… as to keep things in proper balance…

a) error rates are down by a factor of 5-10. (now less than .5% … avg .2%)
b) AMD’s haul butt :slight_smile:
c) MAX WU time has been 30 CPU minutes on a dual channel A-64, few
page faults… very smooth operation.
d) Linux side credits are 1/2 of what they should be… work arounds/fixes
being addressed. Performance equal to Windows.
e) WUs not compressed yet (like Seti)… will be addressed.

Open issues:
a) Completion of the ‘Change of Guard’
b) Completion of new protein target parameter WU testing
c) Adjustments for fair Linux scores
d) Schedule for implementing ‘zipped’ WUs.
e) Transfer of credits for all testing. Not yet guaranteed, but promising.

Rating the current P@H status on a 1-10 scale (1=bad, 10=perfect).

Most things are at a 9-10, a few are at a 5 and a couple are at 0.

Overall Rating, considering a “Change of Guard” AND Startup… 7-8.

When it comes back online, It will haul and be solid. It will also be better
than the P@H we knew before. It has improved that much.

Chuck

6

Nice to see progress, if you don’t want to swamp the servers when they come up just change your preferences so that Predictor is 5% or less - should do it in the short term & no messing with clients

7

Thanks for filling in the news Chuck and good to hear from you :thumbsup:

8

Doesn’t this have to do with using an optimized version of BOINC for their architecture? When I compiled BOINC with optimizations for my tbird, it almost doubled my benchmarks, which in turn doubled the credit requested when submitting a WU (since the WU still takes the same amount of time). before: 26.3 credits requested for 21,660 seconds after: 48 credits for 21,600 seconds

9

It might indeed… I will do a full compile on all platforms and see what is happening. It also might be happening in reverse too… CPU is fast, mem slow.
This coincides with the rest of this post.

Since my original post… I have learned one thing… Suse 9.1 Professional does NOT support the Nvidia Nforce3 Pro 150 MCP (dual channel) memory controller.

Their customer support simply refused to discuss it. Therefore I STRONGLY advise folks to be carefull of what they use.

So this leads to an off-topic question, which is probably best answered via PM… What is the BEST unix/linux for me to use on this board? It HAULS BUTT on everything but suse. I was shocked when native 64 bit math was 1/2 the speed of 32 bit math on windows… and even more shocked when running wine under Suse (32 bit windows compile) was faster than the native compile. It sounds like Suse 9.1 has a few ‘issues’ for Chipsets. I did find that 2.6.10 RC1 of the kernel has PCI ID updates to provide support for the chipset, but that still leaves out the driver. (which explains why my Xfree86 can’t see the AGP sideband, et al correctly and install the driver from ATI).

SO… after all my gyrations… What is a good, simple. STABLE, and (obviously) fast OS for 64 bit, dual channel ECC2 DDR-400… I am about to do something wild (Gnu MP math)… which needs 64 bit

Has FC2 and or LTSP broadened it’s suppport for such things that I need?

OH,… in closing … P@H is, IMHO, ready… The changing of the guard needs to be completed and then we will be online. Everything else is good enough now (based on last few days changes) to go live.

Chuck

PS: it’s great to be back more and more… So, one good final round of testing and we will be P@H’ing again… PPS: I think i have it solid that all credits earned during test will transfer.

10

:thumbsup:

11

ATTENTION ALL!!!

After several emails and evaluations of what is happening with Predictor,
here is the status and plan.


Status:

  1. There is a small validator bug… (possibly). We will not know until a heavier load is placed on the system

  2. The WU generator isn’t creating work fast enough because it is ‘demand based’… and right now, there is no demand for work. It may stay this way during the load test or it may jump into normal mode since it is demand driven.

  3. Since testing began, all but a couple WU’s run clean… There have been 8 failures (WUs) over 1000+ WU’s since testing began.

  4. I, have been authorized to coordinate the last phase of the load testing while the P@H staff is at CASP 6. I am the acting test director until CASP 6 is over or we go live.

  5. Mac’s are still having issues. This will be resolved when the staff returns.

6. THE MESSAGE BOARDS ARE DOWN UNTIL AFTER TESTING AND SWITCH TO LIVE STATUS

Summary: We aren’t putting enough of a load on the system for it to work properly!!! :smiley:


Plan/Action:

I am now accepting requests for the URL and activation of accounts, contingent on confidentiality of the test URL.

The test URL will be changed to the live URL if testing is successfull (which it should be considering the data examined). THEREFORE, ALL POINTS ACCRUED WILL COUNT!!!

I will add 5-10 users at a time, attempting to maintain a balance of Intel/AMD and Windows/Linux. 1-2 CPUs/user is assumed unless you tell me otherwise. I expect to add 5-10/day (avg 20 cpus)

I will pause at 50 users briefly, let it stabilize and we will evaluate… . If the sytem is working better and smoother, we will continue to increase the load up to 100. Should we reach the 100 user milestone, we will evaluate whether to go to 200 or not at that time or remove any CPU restrictions.

Testing is open to all existing P@H users on a first come, first served basis consistent with the priority of AMD/Intel, Windows/Linux balance.

I would like to communicate with each user via a distribution list, so please have an email address that you don’t mind being seen by others. If you need confidentiality, please inform me and I will abide by your wishes and handle it.

So, everyone… Crank’em up… it’s time to have some fun!

Predictor is going to go live … Let’s give it one last load test.

I can be reached currently via my address on file here at TPR. I will create an email address for all to use

When emailing me, I do request you put “Predictor Test Team Signup” in the subject.

'tiz a great day!!!

Let’s have some fun!!!