Workunit size vs. processor

Graham G3ZOD
Graham G3ZOD
Joined: 11 Jul 06
Posts: 12
Credit: 1199477
RAC: 0
Topic 191591

Just out of curiousity ... are specific workunits allocated to computers based on the computers' performance?

Just wondering ... I've been using a puny 800MHz system that's been getting S5 workunits that take about 4 hours on it. I've moved over to a 3.4GHz system and was kinda expecting to rip through workunits, but the first one I got looks to be a juggernaut S5 ... about 14 hours even at 3.4GHz by the looks of things.
Example of baby workunit
Juggernaut workunit

Graham G3ZOD
Graham G3ZOD
Joined: 11 Jul 06
Posts: 12
Credit: 1199477
RAC: 0

Workunit size vs. processor

Okay, I just read the following in another thread, but I'm still curious whether these are being allocated on the basis of the client computers' processing power(?)

Quote:

There are two types of workunits: short and long. The short workunits have XXXX.X less than or equal to 0400.0.

There are also two types of data files: short and long. The short data files (l1_XXXX.X) are from the LIGO Livingston Observatory, and are about 4.5MB in size. The long data files (h1_XXXX.X) are from LIGO Hanford and are about 16MB in size. Note: once your computer downloads one of these data files, it should be able to do many workunits for that same file.

Bruce Allen
Bruce Allen
Moderator
Joined: 15 Oct 04
Posts: 1119
Credit: 172127663
RAC: 0

RE: Just out of curiousity

Quote:

Just out of curiousity ... are specific workunits allocated to computers based on the computers' performance?

Just wondering ... I've been using a puny 800MHz system that's been getting S5 workunits that take about 4 hours on it. I've moved over to a 3.4GHz system and was kinda expecting to rip through workunits, but the first one I got looks to be a juggernaut S5 ... about 14 hours even at 3.4GHz by the looks of things.
Example of baby workunit
Juggernaut workunit

The scheduler is designed to send out short WU to slow hosts, and long WU to fast hosts. The slow WU are sent to the hosts which are in the slowest 20%.

Bruce

Director, Einstein@Home

Graham G3ZOD
Graham G3ZOD
Joined: 11 Jul 06
Posts: 12
Credit: 1199477
RAC: 0

RE: The scheduler is

Message 42653 in response to message 42652

Quote:
The scheduler is designed to send out short WU to slow hosts, and long WU to fast hosts. The slow WU are sent to the hosts which are in the slowest 20%.


Hi Bruce, thanks for clearing that up.

Winterknight
Winterknight
Joined: 4 Jun 05
Posts: 1222
Credit: 312469585
RAC: 646769

RE: RE: Just out of

Message 42654 in response to message 42652

Quote:
Quote:

Just out of curiousity ... are specific workunits allocated to computers based on the computers' performance?

Just wondering ... I've been using a puny 800MHz system that's been getting S5 workunits that take about 4 hours on it. I've moved over to a 3.4GHz system and was kinda expecting to rip through workunits, but the first one I got looks to be a juggernaut S5 ... about 14 hours even at 3.4GHz by the looks of things.
Example of baby workunit
Juggernaut workunit

The scheduler is designed to send out short WU to slow hosts, and long WU to fast hosts. The slow WU are sent to the hosts which are in the slowest 20%.

Bruce


If this is so why did my P3 get long WU and is speeding ahead of the pack, see [url=http://einsteinathome.org/node/191578]Pending: for how long{/url] the pending has now risen to 12 units + 2 WU processing. It means that this computer has received no credit for 10days. RAC dropping fast should be 176.61+ as WU take just under 1 day to crunch.

And my Pent M gets short units, when it does long units in under 10 hrs.

What are the rules that determine work is not been done and the file needs to be sent to other hosts?
I see one unit has, but to a new member i.e. 0 credit. So will it be crunched and if so in what time frame, I (we) just don't know.

Odysseus
Odysseus
Joined: 17 Dec 05
Posts: 372
Credit: 19557274
RAC: 3285

RE: The scheduler is

Message 42655 in response to message 42652

Quote:
The scheduler is designed to send out short WU to slow hosts, and long WU to fast hosts. The slow WU are sent to the hosts which are in the slowest 20%.


I guess it doesn’t always manage: one of my Mac G4/400s has been doing longs (it’s only on its second one) whose names start with l1_1494.0. Not that I’m complaining: the deadlines seem manageable enough even for that system, which isn’t BOINCing 100% of the time and is attached to two other projects.

Barrie
Barrie
Joined: 23 Mar 05
Posts: 219
Credit: 21449106
RAC: 0

Bernd has said in the past

Bernd has said in the past that the strategy for matching WU size to processor speed is not deterministic, but simply bends the probability of getting appropriate WUs in the right direction. No guarantees.

Dead men don't get the baby washed. HTH

Stefan
Stefan
Joined: 15 Nov 05
Posts: 52
Credit: 761198
RAC: 0

RE: Bernd has said in the

Message 42657 in response to message 42656

Quote:
Bernd has said in the past that the strategy for matching WU size to processor speed is not deterministic, but simply bends the probability of getting appropriate WUs in the right direction. No guarantees.

It looks like I'm now getting small WU's, the few long WU's that I sent off must of taken too long and switched mine over to shorter ones...weird though, I've got a p4 2.8 ghz...

And to think, that was fast a year ago ;)

Human Stupidity Is Infinite...

Steve Cressman
Steve Cressman
Joined: 9 Feb 05
Posts: 104
Credit: 139654
RAC: 0

RE: RE: Bernd has said in

Message 42658 in response to message 42657

Quote:
Quote:
Bernd has said in the past that the strategy for matching WU size to processor speed is not deterministic, but simply bends the probability of getting appropriate WUs in the right direction. No guarantees.

It looks like I'm now getting small WU's, the few long WU's that I sent off must of taken too long and switched mine over to shorter ones...weird though, I've got a p4 2.8 ghz...

And to think, that was fast a year ago ;)


Are you sure that yours is running as well as it could? My 4 year old system at 2.1 GHz gets the long ones (178 pointers) done in ~10 hours or less depending on how much i use it for other stuff.

98SE XP2500+ @ 2.1 GHz Boinc v5.8.8

Bird-Dog
Bird-Dog
Joined: 5 Nov 05
Posts: 617
Credit: 5824655
RAC: 0

RE: RE: RE: Bernd has

Message 42659 in response to message 42658

Quote:
Quote:
Quote:
Bernd has said in the past that the strategy for matching WU size to processor speed is not deterministic, but simply bends the probability of getting appropriate WUs in the right direction. No guarantees.

It looks like I'm now getting small WU's, the few long WU's that I sent off must of taken too long and switched mine over to shorter ones...weird though, I've got a p4 2.8 ghz...

And to think, that was fast a year ago ;)


Are you sure that yours is running as well as it could? My 4 year old system at 2.1 GHz gets the long ones (178 pointers) done in ~10 hours or less depending on how much i use it for other stuff.

AMD`s seem to be much quicker on this project. I can return a result in about 10hrs on a 2ghz AMD Athlon, much quicker than a 3.2ghz p4 I`ve been paired with.

archae86
archae86
Joined: 6 Dec 05
Posts: 3145
Credit: 7024914931
RAC: 1808801

RE: AMD`s seem to be much

Message 42660 in response to message 42659

Quote:

AMD`s seem to be much quicker on this project. I can return a result in about 10hrs on a 2ghz AMD Athlon, much quicker than a 3.2ghz p4 I`ve been paired with.

Umm... did you notice that most of your P4 partners are running hyperthreaded? I did not click enough times to find the 3.2 rig, but saw a 2.8 rig which when you divided the reported CPU time by two to account give credit for the work done on the other thread appeared to give a productivity better than yours, with a productivity per Hz slightly below yours (though that is a meaningless number).

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.