climateprediction.net home page
climateprediction slow on linux boxes?

climateprediction slow on linux boxes?

Questions and Answers : Unix/Linux : climateprediction slow on linux boxes?
Message board moderation

To post messages, you must log in.

AuthorMessage
Scarf

Send message
Joined: 25 Aug 04
Posts: 9
Credit: 11,258,003
RAC: 0
Message 8272 - Posted: 31 Jan 2005, 20:22:33 UTC

When I compared the climateprediction benchmark results of my 2400+ with Windows XP and my 3000+ with Debian Linux Sarge with 2.6.10 Kernel, i wondered why the Windows machine is about twice as fast as the Linux box.

results:

3000+ Linux:
Measured floating point speed 1091 million ops/sec
Measured integer speed 2279.2 million ops/sec

2400+@2176 MHz Windows:
Measured floating point speed 2003.3 million ops/sec
Measured integer speed 4723.4 million ops/sec

is this a general issue of the climateprediction linux client or is there some other reason (eg. preemptive kernel)?
ID: 8272 · Report as offensive     Reply Quote
Profile geophi
Volunteer moderator

Send message
Joined: 7 Aug 04
Posts: 1929
Credit: 41,067,747
RAC: 14,699
Message 8275 - Posted: 31 Jan 2005, 20:47:21 UTC
Last modified: 31 Jan 2005, 20:47:55 UTC

This is a BOINC issue. The benchmarks are meaningless in CPDN performance and credit. With your system, the Linux and Windows compiled versions of CPDN should perform very similarly on the same PC.
ID: 8275 · Report as offensive     Reply Quote
Profile Andrew Hingston
Volunteer moderator

Send message
Joined: 17 Aug 04
Posts: 753
Credit: 9,804,700
RAC: 0
Message 8293 - Posted: 1 Feb 2005, 0:07:45 UTC

On my 2.66 GHz P4 box, CPDN runs about 10% faster under SUSE 9.2 than under WinXP.
ID: 8293 · Report as offensive     Reply Quote
old_user30025

Send message
Joined: 15 Nov 04
Posts: 19
Credit: 35,499
RAC: 0
Message 8731 - Posted: 6 Feb 2005, 12:46:42 UTC

Yes, I also get very weird numbers for my Linux computers with respect to ALL numbers (uptime, up/download rate, % online, % working on the WU, time per TS, etc.). I think it is safe to assume they do not mean a thing, or do they?
ID: 8731 · Report as offensive     Reply Quote
Profile old_user248

Send message
Joined: 6 Aug 04
Posts: 65
Credit: 1,605,224
RAC: 0
Message 8736 - Posted: 6 Feb 2005, 12:58:21 UTC - in response to Message 8731.  

> Yes, I also get very weird numbers for my Linux computers with respect to ALL
> numbers (uptime, up/download rate,....

For my dual Xeon EMT64 3.2 Ghz @ 3.456 GHz running FC3 & 4 models I get FPU of 738.4 and Integer of 1555 which is worse than all my machines, including a 2 GHz laptop running XPPro, so Yup, numbers kinda meaningless.

I haven't done a speed comparison, which is very difficult due to processing rates of different models but concensus is that things run faster on Linux.
ID: 8736 · Report as offensive     Reply Quote
Profile old_user60427
Avatar

Send message
Joined: 4 Mar 05
Posts: 24
Credit: 243,647
RAC: 0
Message 11013 - Posted: 16 Mar 2005, 20:22:53 UTC - in response to Message 8275.  

> This is a BOINC issue. The benchmarks are meaningless in CPDN performance and
> credit. With your system, the Linux and Windows compiled versions of CPDN
> should perform very similarly on the same PC.
>

I am aware of the limited value of the BOINC benchmarks -- use an optimised BOINC executable myself (see http://www.pperry.f2s.com/index.htm or http://boinc.us.tt/) to minimse the issue a bit.

I know that for SETI there is no difference between linux & windows --both science apps will run approx as fast on the same hardware. I did however just found out that the Einstein app is only running at ~50% of the speed under Windows, presumably because the math gets optimised much more under windows (better compiler).

I am looking for another project (next to my main project, SETI), like the idea of CPDN, can probably manage the HD requirement, but am wondering how performance will compare across platforms (using one Windows & one Linux PC).
ID: 11013 · Report as offensive     Reply Quote
Profile geophi
Volunteer moderator

Send message
Joined: 7 Aug 04
Posts: 1929
Credit: 41,067,747
RAC: 14,699
Message 11014 - Posted: 16 Mar 2005, 21:29:42 UTC - in response to Message 11013.  

> I am looking for another project (next to my main project, SETI), like the
> idea of CPDN, can probably manage the HD requirement, but am wondering how
> performance will compare across platforms (using one Windows & one Linux
> PC).
>
On an AMD CPU, there is little or no difference running Linux vs. Windows. At least that was my experience on an Athlon64 3200+. However, on Intel CPUs, and P4s in particular, Linux is noticeably faster. How much? Maybe 10 to 20 percent at least for this first experiment with the slab model.

Some of us are alpha testing a sulphur experiment model, and they have not been able to get it stable on Linux with the same type of optimizations as Windows, so Linux is considerably slower than Windows, so far in that model.
ID: 11014 · Report as offensive     Reply Quote

Questions and Answers : Unix/Linux : climateprediction slow on linux boxes?

©2020 climateprediction.net