Index  | Recent Threads  | Unanswered Threads  | Who's Active  | Guidelines  | Search
 

Quick Go »
No member browsing this thread
Thread Status: Active
Total posts in this thread: 9
[ Jump to Last Post ]
Post new Thread
Author
Previous Thread This topic has been viewed 1862 times and has 8 replies Next Thread
Hypernova
Master Cruncher
Audaces Fortuna Juvat ! Vaud - Switzerland
Joined: Dec 16, 2008
Post Count: 1908
Status: Offline
Project Badges:
Reply to this Post  Reply with Quote 
[RESOLVED] Crunching Efficiency

If we look at the Points per Hour of Runtime criteria we see that DDDT2 has the double number of points than FAAH. shock

Why such a big difference. What underlying mechanism creates such a discrepancy? Is it the types of calculations. Maybe in the same timeframe DDDT2 crunches more and FAAH does more disk access or file transfers etc.

It would be a way to estimate the crunching efficiency of a project.

Any ideas? thinking
----------------------------------------

----------------------------------------
[Edit 1 times, last edit by Hypernova at May 29, 2010 3:02:44 PM]
[May 26, 2010 1:15:28 PM]   Link   Report threatening or abusive post: please login first  Go to top 
Former Member
Cruncher
Joined: May 22, 2018
Post Count: 0
Status: Offline
Reply to this Post  Reply with Quote 
Re: Crunching Efficiency

Perhaps part of this discrepancy is due to AIDS having been running here since 2005 and DDDT2 having been started recently. Five years ago most computers were much slower than those we have available today, and therefore generating fewer point per hour.
[May 26, 2010 1:32:46 PM]   Link   Report threatening or abusive post: please login first  Go to top 
KerSamson
Master Cruncher
Switzerland
Joined: Jan 29, 2007
Post Count: 1684
Status: Offline
Project Badges:
Reply to this Post  Reply with Quote 
Re: Crunching Efficiency

Hallo Hypernova,
some slight differences can exist between projetcs. In the case you mention, I am very surprised by such a big difference. For this reason, I have the following questions:
  • Are you sure that the difference is not the result of a measurement artefact?
  • I notice that your hosts are mostly based on i7-Exterme CPU: does this CPU have a specific weakness by some computation mechanisms intensively used by FA@H?

Especially regarding the second questions, it would be very interesting to check if other members can observe the same effect with FA@H on different CPUs.
Normally, the age of a project does not have an impact on the "credit efficiency". The various project reports generated by Sekerob (see his graphics) can widely confirm this statement.
Cheers,
Yves
----------------------------------------
[May 26, 2010 10:31:35 PM]   Link   Report threatening or abusive post: please login first  Go to top 
Former Member
Cruncher
Joined: May 22, 2018
Post Count: 0
Status: Offline
Reply to this Post  Reply with Quote 
Re: Crunching Efficiency

Points are based on Flops completed, not time. the older projects "Average" time per work unit is higher than the newer becuase of increases in cpu speed, so more flops/sec = less time = more points per sec, for the same number of flops. also DDDT2 is limited to faster machines, make the differance appear even greater.

compare both projects on the same box and I suspect you will get similar points/hr numbers
[May 26, 2010 11:21:00 PM]   Link   Report threatening or abusive post: please login first  Go to top 
Hypernova
Master Cruncher
Audaces Fortuna Juvat ! Vaud - Switzerland
Joined: Dec 16, 2008
Post Count: 1908
Status: Offline
Project Badges:
Reply to this Post  Reply with Quote 
Re: Crunching Efficiency

Hallo Hypernova,
some slight differences can exist between projetcs. In the case you mention, I am very surprised by such a big difference. For this reason, I have the following questions:
  • Are you sure that the difference is not the result of a measurement artefact?
  • I notice that your hosts are mostly based on i7-Exterme CPU: does this CPU have a specific weakness by some computation mechanisms intensively used by FA@H?

Especially regarding the second questions, it would be very interesting to check if other members can observe the same effect with FA@H on different CPUs.
Normally, the age of a project does not have an impact on the "credit efficiency". The various project reports generated by Sekerob (see his graphics) can widely confirm this statement.
Cheers,
Yves



Yves, In fact I did not measure these values.
I took the values as per Project Statistics where you have for each project the average values for Runtime/WU, Points/Hour etc.
These are WCG "official average project numbers".
This means that these are values averaged over an extremely large set of different computers. So hardware is not relevant here, hence my question on the software.
----------------------------------------

[May 27, 2010 8:17:47 AM]   Link   Report threatening or abusive post: please login first  Go to top 
KerSamson
Master Cruncher
Switzerland
Joined: Jan 29, 2007
Post Count: 1684
Status: Offline
Project Badges:
Reply to this Post  Reply with Quote 
Re: Crunching Efficiency

Hi Hypernova,
compiling fredski's answer and the details your provided, the answer is simple !
3 or 5 years ago, CPUs were less powerful and they took more time for crunching WUs. The simple average calculation you did, based on the WCG's overall statistics, shows clearly this evolution.
You should perform the same average calculation based on a 1 week or 1 month delta, and you would probably notice that the differences between projects are less than 10%.
Cheers,
Yves
----------------------------------------
[May 27, 2010 11:49:17 AM]   Link   Report threatening or abusive post: please login first  Go to top 
Former Member
Cruncher
Joined: May 22, 2018
Post Count: 0
Status: Offline
Reply to this Post  Reply with Quote 
Re: Crunching Efficiency

As a measure of project crunching efficiency, I have calculated the points per hour for yesterday (26 May 2010) from the WCG Global Statistics and By Projects pages. DDDT2 is not included because there were no results for yesterday.

One day is a small sample, but it includes over 2.6 million hours (300 years) and 614,936 results.

Project Points/Hour
---------------------
HCMD2 109.31
HFCC 100.53
HCC 102.34
HPF2 92.98
AIDS 102.26

WCG 101.92
[May 27, 2010 1:53:15 PM]   Link   Report threatening or abusive post: please login first  Go to top 
sk..
Master Cruncher
http://s17.rimg.info/ccb5d62bd3e856cc0d1df9b0ee2f7f6a.gif
Joined: Mar 22, 2007
Post Count: 2324
Status: Offline
Project Badges:
Reply to this Post  Reply with Quote 
Re: Crunching Efficiency

Steve, nice set of results. Thanks.

Some people only have high end i7 computers, and some only have dual cores, so the average computer may not accurately reflect personal findings.
I found at least 10% differences in terms of points per project for some systems, including one low end i7, if there is such a thing?
I tested a variety of systems and each was different in some performance way, but generally I think it is fair to say that the more recent tasks perform better on more recent CPUs.
Everyone should note that in terms of points, it is better to crunch at least 2 projects. It's just finding the two or more that is the hard part, and it's very-very hard to work out, for each system, especially given the fact that project WU's change, sometimes quite often (Aids)!
[May 28, 2010 9:30:52 PM]   Link   Report threatening or abusive post: please login first  Go to top 
KerSamson
Master Cruncher
Switzerland
Joined: Jan 29, 2007
Post Count: 1684
Status: Offline
Project Badges:
Reply to this Post  Reply with Quote 
Re: Crunching Efficiency

Hello Steveleg,
thank you for the short statistical review.
Indeed the real results could vary depending of the host configuration.
After several months as HCMD2, I made some weeks ago few weeks for HFCC. For my surprise, the average daily credits were around 5-7% higher than for HCMD2.
But I do not have any i7 and the major part of my hosts based on the "old" core2 (Quad) or similar Xeon architecture (3 years old).
Finally, your review confirm that WCG is really consistent regarding granted credits.
Cheers,
Yves
----------------------------------------
[May 29, 2010 11:45:02 AM]   Link   Report threatening or abusive post: please login first  Go to top 
[ Jump to Last Post ]
Post new Thread