World Community Grid - View Thread - Long Projected Completion Times for Genome Comparison

World Community Grid Forums

Category: Completed Research

Forum: Genome Comparison

Thread: Long Projected Completion Times for Genome Comparison

Quick Go »

No member browsing this thread

Thread Status: Active
Total posts in this thread: 6

[ ]

Author

This topic has been viewed 2498 times and has 5 replies

Sekerob
Ace Cruncher
Joined: Jul 24, 2005
Post Count: 20043
Status: Offline


Long Projected Completion Times for Genome Comparison

Though the box is racing thru them, the project completion times in the Task Buffer, seem to be going up rather than down.... 2:17 hours actual achieved (longest seen) v 6:52 projected... when the first came thru it was about 6:15.

It only seems to effect the GC. The HDC which for a period were quite long, dropped rapidly in projected times during the same period. The Duration Correction Factor in the state file only slightly increased from a week ago to 0.80 i.e. still processing WU's substantially faster than the benchmark would indicate.

Anyone else observing these excessive projection times and depletion of the work/task buffer?

----------------------------------------

WCG

Global & Research > Make Proposal Help: Start Here!
Please help to make the Forums an enjoyable experience for All!

[Nov 21, 2006 2:18:53 PM]

Former Member
Cruncher
Joined: May 22, 2018
Post Count: 0
Status: Offline


Re: Long Projected Completion Times for Genome Comparison

I'm not seeing such a pronounced error, but it does look as if the flop estimation being sent with the work units is off a little. The error will correct itself for anyone running just GC, but anyone with a mix of projects will get badly wrong time estimation.

I think this is something Kevin knows how to fix, since we had a similar problem with FAAH.

[Nov 21, 2006 2:46:48 PM]

Former Member
Cruncher
Joined: May 22, 2018
Post Count: 0
Status: Offline


Re: Long Projected Completion Times for Genome Comparison

I've checked the estimations, and this is what I found (reformated for readability):

<workunit>
    <name>10000066-10000101</name>
    <app_name>fcg1</app_name>
    <version_num>510</version_num>
    <rsc_fpops_est>30,000,000,000,000</rsc_fpops_est>
    <rsc_fpops_bound>510,870,220,560,000</rsc_fpops_bound>
    <rsc_memory_bound>15,000,000</rsc_memory_bound>
    <rsc_disk_bound>10,000,000</rsc_disk_bound>
</workunit>
<workunit>
    <name>faah0983_d132n162_x1AJX_00</name>
    <app_name>faah</app_name>
    <version_num>528</version_num>
    <rsc_fpops_est>45,000,000,000,000</rsc_fpops_est>
    <rsc_fpops_bound>510,870,220,560,000</rsc_fpops_bound>
    <rsc_memory_bound>100,000,000</rsc_memory_bound>
    <rsc_disk_bound>75,000,000</rsc_disk_bound>
</workunit>

As you can see, the bounds estimations are just fine, but the fpops_est is a long way from reality. A GC unit is not 2/3 the size of a FAAH unit - on most computers it takes less than half the time, and on my slowest computer that's more like a quarter. WCG should hopefully have some better figures now the project has been running for a while.

[Nov 22, 2006 7:01:17 PM]

Sekerob
Ace Cruncher
Joined: Jul 24, 2005
Post Count: 20043
Status: Offline


Re: Long Projected Completion Times for Genome Comparison

Confirmed.... 1/4 is about the achieved time for GC v FAAH on my box.

----------------------------------------

WCG

Global & Research > Make Proposal Help: Start Here!
Please help to make the Forums an enjoyable experience for All!

[Nov 22, 2006 7:07:42 PM]

Former Member
Cruncher
Joined: May 22, 2018
Post Count: 0
Status: Offline


Re: Long Projected Completion Times for Genome Comparison

Is this what your talking about. See the highlighted one?

[Nov 22, 2006 7:21:01 PM]

Sekerob
Ace Cruncher
Joined: Jul 24, 2005
Post Count: 20043
Status: Offline


Re: Long Projected Completion Times for Genome Comparison

Brinktastee,

Right on the button... looks like your other computers have learned, but BRINK-PUTER not yet (lest it's a P3 ;>). After about 20 GCs here, the projection is now down to 4.5 hours which is still double the actual. As laid out by Didactylos, the flops setting in the transmitted file are way to high. Based on the individual machines Benchmark, the instance will estimate the time.

You got a serious collection there in the BOiNCview applause

----------------------------------------

WCG

Global & Research > Make Proposal Help: Start Here!
Please help to make the Forums an enjoyable experience for All!

[Nov 22, 2006 8:01:49 PM]

[ ]