I read paper comparing performance of L4Linux of MkLinux (atop OSFMK) on x86 computer. Results are very interesting. I would really like to conduct benchmarks on my own. I know where to download Linux and L4Linux, but not the MkLinux for Intel. Does anyone still have binaries? If you do, please let me know how I could get these. Thank you, IS.
I read paper comparing performance of L4Linux of MkLinux (atop OSFMK) on x86 computer. Results are very interesting. I would really like to conduct benchmarks on my own. I know where to download Linux and L4Linux, but not the MkLinux for Intel. Does anyone still have binaries? If you do, please let me know how I could get these. Thank you, IS.
You may want to have a look at http://www.mklinux.org/ I compiled MkLinux and OSF/Mk from the CVS sources on Intel a while ago. Don't know if it is still possible... Regards, -Farid. -- Farid Hajji -- Unix Systems and Network Admin | Phone: +49-2131-67-555 Broicherdorfstr. 83, D-41564 Kaarst, Germany | farid.hajji@ob.kamp.net - - - - - - - - - - - - - - - - - - - - - - - + - - - - - - - - - - - - Due to budget cuts, light at end of tunnel will be out. --Unknown.
You may want to have a look at http://www.mklinux.org/ They don't have x86 stuff.
I compiled MkLinux and OSF/Mk from the CVS sources on Intel a while ago. Don't know if it is still possible... Not with modern gcc and not sure if working x86 binaries could be produced from unaltered sources.
So if someone still has linux server, please either email me compressed binary or let me know what to do, to obtain it. Sincerely, IS.
You may want to have a look at http://www.mklinux.org/ They don't have x86 stuff. Yes, I know :-(
I compiled MkLinux and OSF/Mk from the CVS sources on Intel a while ago. Don't know if it is still possible... Not with modern gcc and not sure if working x86 binaries could be produced from unaltered sources. Last time I compiled and ran both osfmk and mklinux on x86 was approx. 20 months ago (under FreeBSD).
Sources didn't compile out of the box. I had to fiddle with the compile tools pretty heavily back then. Yes, it will be difficult to back out their recent changes...
So if someone still has linux server, please either email me compressed binary or let me know what to do, to obtain it. Sorry, I don't have the binaries anymore.
Good luck! -Farid. -- Farid Hajji -- Unix Systems and Network Admin | Phone: +49-2131-67-555 Broicherdorfstr. 83, D-41564 Kaarst, Germany | farid.hajji@ob.kamp.net - - - - - - - - - - - - - - - - - - - - - - - + - - - - - - - - - - - - Due to budget cuts, light at end of tunnel will be out. --Unknown.
[Farid Hajji]
So if someone still has linux server, please either email me compressed binary or let me know what to do, to obtain it.
Sorry, I don't have the binaries anymore.
Correct me if I'm wrong, but wasn't MkLinux using a kernel from the 2.0 series? If so, you should remember that comparing with a current L4Linux will be like comparing apples and oranges (depending on what you intend to measure, that is). eSk
Correct me if I'm wrong, but wasn't MkLinux using a kernel from the 2.0 series? Correct.
If so, you should remember that comparing with a current L4Linux will be like comparing apples and oranges (depending on what you intend to measure, that is). You're absolutely right.
eSk
-Farid. -- Farid Hajji -- Unix Systems and Network Admin | Phone: +49-2131-67-555 Broicherdorfstr. 83, D-41564 Kaarst, Germany | farid.hajji@ob.kamp.net - - - - - - - - - - - - - - - - - - - - - - - + - - - - - - - - - - - - Due to budget cuts, light at end of tunnel will be out. --Unknown.
If so, you should remember that comparing with a current L4Linux will be like comparing apples and oranges (depending on what you intend to measure, that is). You're absolutely right.
Yes, comparing modern l4linux and 2.0 mklinux won't yield too much. I already replied to author, but since you sent copy to the list, I'll reply to list as well. There is a paper which compares MkLinux vs. L4Linux. Pager is quite old (about 5 years AFAIR), however ... I remember reading in that very paper that - no proof that Mach based systems give only about 5% degradation was found, thus it is not being concedered a valid statement. Then comparison benchmarks are being given (though never analyzed). Extending same logic, I could say following: Paper is out, but due to lack of independently verified results and/or infrustrure required to complete such evaluation, results should not seen as valid. Thus, IMHO currently paper is of interest as a sample usenix submition, but not a research report. (until results could be duplicated, which requires working x86 binaries of both l4linux and mklinux running 2.0 kernels) Farid said that he at some point compiled (more than once) mklinux for x86. Perhaps, would be worth doing it again, otherwise often seen claim that l4 is better microkernel, is more like rant. I did not mean to offend anyone, but just trying to see which claims are valid and which are just wishful thinking. Sincerely, IS.
"Igor Shmukler" <shmukler@mail.ru> writes:
I remember reading in that very paper that - no proof that Mach based systems give only about 5% degradation was found, thus it is not being concedered a valid statement.
Do you mean this statement: "We found no substantiation for the ``common knowledge'' that early Mach3.0-based Unix single-server implementations achieved a performance penalty of only 10% compared to bare Unix on the same hardware. For newer hardware, [9] reports penalties of about 50%." (related work chapter of "The Performance of micro-Kernel-Based Systems", http://os.inf.tu-dresden.de/pubs/sosp97/) The Problem was that reviewers of the papers pointed out that Mach-based Unix-single-servers have a penalty of 10% without providing any reference to a conference paper. So we asked around, contacted people who had worked on Mach, but nobody was able to confirm the 10% penalty. Therefore we added this statement to the related work section.
Then comparison benchmarks are being given (though never analyzed). Extending same logic, I could say following: Paper is out, but due to lack of independently verified results and/or infrustrure required to complete such evaluation, results should not seen as valid.
We would have been happy to provide anyone with a copy of our test environment if anybody would have asked for it. And it isn't the same logic. The problem was that some people said, that Mach based Unix-Servers had only 10% performance penalty without providing us with any performance numbers or references to papers. In contrast we compared two micro kernel based implementations of Linux running on the same hardware and published the results.
Farid said that he at some point compiled (more than once) mklinux for x86. Perhaps, would be worth doing it again, otherwise often seen claim that l4 is better microkernel, is more like rant.
Does that mean, you consider the published results as invalid? That we published numbers we didn't measure? At least the cited paper of M. Condict et al confirms our aim benchmark results.
I did not mean to offend anyone, but just trying to see which claims are valid and which are just wishful thinking.
Hope I was able to answer some of your questions. regards, Jean
Do you mean this statement:
"We found no substantiation for the ``common knowledge'' that early Mach3.0-based Unix single-server implementations achieved a performance penalty of only 10% compared to bare Unix on the same hardware. For newer hardware, [9] reports penalties of about 50%." (related work chapter of "The Performance of micro-Kernel-Based Systems", http://os.inf.tu-dresden.de/pubs/sosp97/) Yes, that is what I was talking about.
The Problem was that reviewers of the papers pointed out that Mach-based Unix-single-servers have a penalty of 10% without providing any reference to a conference paper. So we asked around, contacted people who had worked on Mach, but nobody was able to confirm the 10% penalty. Therefore we added this statement to the related work section. I read reports with even lower than 10% penalty results.
We would have been happy to provide anyone with a copy of our test environment if anybody would have asked for it. I am asking for it now.
And it isn't the same logic. The problem was that some people said, that Mach based Unix-Servers had only 10% performance penalty without providing us with any performance numbers or references to papers. In contrast we compared two micro kernel based implementations of Linux running on the same hardware and published the results. For results to be valid, they need to be consistant (reproducable)
Farid said that he at some point compiled (more than once) mklinux for x86. Perhaps, would be worth doing it again, otherwise often seen claim that l4 is better microkernel, is more like rant.
Does that mean, you consider the published results as invalid? That we published numbers we didn't measure? At least the cited paper of M. Condict et al confirms our aim benchmark results. I did not say results are invalid. All I was asking for - binaries to be able to reproduce results. If results are in fact reproducable they are valid. I have no doubts that results MAY BE valid. However, I think that not all conclusions of that paper are valid. If I get a chance to validate results, I would be happy to write full report that would include benchmark results, analisys and conclusion that I will come to.
Hope I was able to answer some of your questions. You probably gave me complete answer, I am just having hard time understanding it. Could I today (or in a near future) get x86 mklinux binaries in order to validate your results?
Thank you, IS.
participants (4)
-
Espen Skoglund -
Farid Hajji -
Igor Shmukler -
Jean Wolter