2.2 and 2.4 performance issues
I was asked (i'm a company freshman) to perform some tests between kernel versions 2.2 and 2.4, and after awhile searching i found a good set of benchmarking tools (aim7) from Caldera linux. I've tested both 2.2.20 and 2.4.18 (preemptive patch) versions in a PIII , and used the default benchmark mixes that already came with the suite. For my great surprise i started having better results with kernel 2.2, only in the DataBase test the kernel 2.4 had better results in everything . I know this might be old news, but i'm also new to this ml. I also read Linus explanation about the handling of the SSE2 signal stack in 2.4, my question is... is there anymore reasons besides this one for the performance downgrade? Are there better way to perform the test besides using benchmark tools like this?
thanx in advance
Quoted 3 lines · Luis Pedro de Moura Ribeiro Pinto
I was asked (i'm a company freshman) to perform some tests between kernel versions 2.2 and 2.4, and after awhile searching i found a good set of benchmarking tools (aim7) from Caldera linux.
Benchmarks are evil. Sure they are useful at times, but for the most part they get misused. IMHO, aim7 is outdated. The I/O it does it all very small for today's systems. It's like poking the system with hundreds of needles. You have no idea how the system will react to a golf club, baseball bat, sledgehammer or a wet noodle. Sure, some people really like it and swear by it. Benchmarking is better done with an application set in mind and best done with the application set itself.
Quoted 2 lines
Are there better way to perform the test besides using benchmark tools like this?
Run the applications you really care about. There is also a good set of benchmarks, including application specific ones at
http://lbs.sourceforge.net/.
Signature · 3 lines
Nate Straz [email protected] sgi, inc http://www.sgi.com/ Linux Test Project http://ltp.sf.net/
Quoted 7 lines · Nathan Straz
Quoted 3 lines · Luis Pedro de Moura Ribeiro Pinto
I was asked (i'm a company freshman) to perform some tests between kernel versions 2.2 and 2.4, and after awhile searching i found a good set of benchmarking tools (aim7) from Caldera linux.
Benchmarks are evil. Sure they are useful at times, but for the most part they get misused.
There's no sense denying evidence that 2.2 outperforms 2.4 under certain workloads. Instead we should just be more determined to root out all those problems and deal with them. There is no inherent design reason why 2.4 should be slower than 2.2 in any area at all, however, some practical issues, such as IO scheduling still remain and are being actively worked on. Expect backports from 2.5 later in the 2.4 series. For now, the one thing we must not do is risk instability in 2.4, now that most users have switched over to it.
Signature · 1 line
Daniel
Quoted 8 lines · Daniel Phillips
Quoted 1 line · Nathan Straz
On Fri, Jun 21, 2002 at 05:55:55PM +0100, Luis Pedro de Moura Ribeiro Pinto
Quoted 4 lines · wrote:
Quoted 3 lines
I was asked (i'm a company freshman) to perform some tests between kernel versions 2.2 and 2.4, and after awhile searching i found a good set of benchmarking tools (aim7) from Caldera linux.
Quoted 7 lines
Quoted 2 lines
Benchmarks are evil. Sure they are useful at times, but for the most part they get misused.
There's no sense denying evidence that 2.2 outperforms 2.4 under certain workloads. Instead we should just be more determined to root out all those problems and deal with them. There is no inherent design reason why 2.4 should be slower than 2.2 in any area at all,
As mentioned earlier in 2.4 with SSE compiled in the context switch costs will be slightly higher. So unless this expense can be made up in another area, or we can hide it with laziness...
But in general I agree.
Quoted 3 lines
however, some
practical issues, such as IO scheduling still remain and are being actively worked on.
Yep always more work todo :)
Quoted 3 lines
Expect backports from 2.5 later in the 2.4 series.
For now, the one thing we must not do is risk instability in 2.4, now that most users have switched over to it.
Definitely.
Eric