Re: macosx 1 process, linux N processes, macosx chokes under load

This WebDNA talk-list message is from

2001

It keeps the original formatting. numero = 36827
interpreted = N
texte = Dale, From a mac rumor site:....But most importantly, Apple has been hard at work in the deepest innerregions of the operating system working to improve the way the Mach kernel'stask manager handles application events (a particular problem for Carbonapplications presently, we're told) and higher levels of the core OS codeare being audited line-by-line for performance speedbumps. Now that Applehas fewer bugs and essential user-interface issues to address compared towhen it was hard at work on 10.0, far more attention is being focused onperformance....and it shows......I would say Apple knows about these problems and is working on it - we allknow OSX sometimes goes to sleep and seems to get stuck on things. Hopefullywe will see a big boost in performance with 10.1 (the REAL release of OSX)Have you done the tests on an OSX Server 2.0 machine ?? How did thatcompare.Alain> Le Pham wrote:>> Dale LaFountain wrote:>>> >>> I have been doing extensive performance testing to find the fastest>>> hardware/software combination to run our websites. In my testing, I>>> found that for single executions, MacOSX executes webcat code about>>> 33% faster than a similar Linux box (P3-550 dual).>>> >>> However, when I execute any number of simultaneous requests using a>>> performance tool like Netsplat, MacOSX quickly grinds to a halt while>>> Linux keeps up rather well.>>> >>> Upon further inspection, it appears that Linux webcat starts 4>>> processes on launch, and then spawns additional child processes for>>> every simultanous request. The OSX version runs ALL requests through>>> a single process, which is apparently a huge bottleneck.>> FYI, Mac OS X does spwan threads to handle request, it is just in>> different way. Use command top during your test you can see>> number of threads that spawned by WebCatalog>> After WebCatalog is launched on Mac OS X, using top command you>> will see there are three threads attached with WebCatalog process>> Linux shows 4 processes. These are include 1 for main WebCatalog>> process and the remain three are spawned threads> > OK, fair enough. I see the threadcount rising now when I run Netsplat on OSX.> >>> >>> Why is webcat set up this way on OSX, and can it be fixed to act more>>> like Linux?>> WebCat itself is using the same method of spwaning threads to handle>> requests. But POSIX thread package is implemented different on>> Linux and Mac OSX. And it is the part of kernel, we could not do>> any thing to force both OS act as the same way.>> >>> >>> Also, the 10.0.4 update to OSXS 2.0 resulted in a 50% improvement in>>> webcat performance in cpu-intensive calculations. However, it's>>> still an order of magnitude slower than Linux when under a heavy load>>> (134 seconds vs 18 seconds).>> We are experience the same result here. WebCat/Linux always runs>> faster the WebCat/Mac OS X Server/Client>> I think it is OS dependent> > OSX is NOT always slower. Running a single instance of my template,> OSX is faster by 50%. It's only when multiple requests occur> simultaneously that OSX crawls, even in comparison to OS9/Webstar on> the same hardware.> > Isn't it possible that some inefficiencies exist in the way> webcat/osx handles threads that could cause this massive difference> in performance? Do other applications that run on both OSX and Linux> have this problem, or is this a problem specific to webcat? If not> in webcat, then the problem should be reported to the Darwin group> and to Apple as a bug.> > If this is an inherent flaw in (current) OSX, I would strongly> discourage anyone from using webcat/osx for busy webcat-driven sites.> It certainly can't keep up with our (relatively simple) testing suite> using Netsplat.> > -Dale-------------------------------------------------------------This message is sent to you because you are subscribed to the mailing list .To unsubscribe, E-mail to: To switch to the DIGEST mode, E-mail to Web Archive of this list is at: http://search.smithmicro.com/ Associated Messages, from the most recent to the oldest:

Re: macosx 1 process, linux N processes, macosx chokes under load (Aaron Lynch 2001)
Re: macosx 1 process, linux N processes, macosx chokes under load (Dale LaFountain 2001)
Re: macosx 1 process, linux N processes, macosx chokes under load (Alain Russell 2001)
Re: macosx 1 process, linux N processes, macosx chokes under load (Aaron Lynch 2001)
Re: macosx 1 process, linux N processes, macosx chokes under load (Dale LaFountain 2001)
Re: macosx 1 process, linux N processes, macosx chokes under load (Aaron Lynch 2001)
Re: macosx 1 process, linux N processes, macosx chokes under load (Alain Russell 2001)
Re: macosx 1 process, linux N processes, macosx chokes under load (Dale LaFountain 2001)
Re: macosx 1 process, linux N processes, macosx chokes under load (Le Pham 2001)
macosx 1 process, linux N processes, macosx chokes under load (Dale LaFountain 2001)

Dale, From a mac rumor site:....But most importantly, Apple has been hard at work in the deepest innerregions of the operating system working to improve the way the Mach kernel'stask manager handles application events (a particular problem for Carbonapplications presently, we're told) and higher levels of the core OS codeare being audited line-by-line for performance speedbumps. Now that Applehas fewer bugs and essential user-interface issues to address compared towhen it was hard at work on 10.0, far more attention is being focused onperformance....and it shows......I would say Apple knows about these problems and is working on it - we allknow OSX sometimes goes to sleep and seems to get stuck on things. Hopefullywe will see a big boost in performance with 10.1 (the REAL release of OSX)Have you done the tests on an OSX Server 2.0 machine ?? How did thatcompare.Alain> Le Pham wrote:>> Dale LaFountain wrote:>>> >>> I have been doing extensive performance testing to find the fastest>>> hardware/software combination to run our websites. In my testing, I>>> found that for single executions, MacOSX executes webcat code about>>> 33% faster than a similar Linux box (P3-550 dual).>>> >>> However, when I execute any number of simultaneous requests using a>>> performance tool like Netsplat, MacOSX quickly grinds to a halt while>>> Linux keeps up rather well.>>> >>> Upon further inspection, it appears that Linux webcat starts 4>>> processes on launch, and then spawns additional child processes for>>> every simultanous request. The OSX version runs ALL requests through>>> a single process, which is apparently a huge bottleneck.>> FYI, Mac OS X does spwan threads to handle request, it is just in>> different way. Use command top during your test you can see>> number of threads that spawned by WebCatalog>> After WebCatalog is launched on Mac OS X, using top command you>> will see there are three threads attached with WebCatalog process>> Linux shows 4 processes. These are include 1 for main WebCatalog>> process and the remain three are spawned threads> > OK, fair enough. I see the threadcount rising now when I run Netsplat on OSX.> >>> >>> Why is webcat set up this way on OSX, and can it be fixed to act more>>> like Linux?>> WebCat itself is using the same method of spwaning threads to handle>> requests. But POSIX thread package is implemented different on>> Linux and Mac OSX. And it is the part of kernel, we could not do>> any thing to force both OS act as the same way.>> >>> >>> Also, the 10.0.4 update to OSXS 2.0 resulted in a 50% improvement in>>> webcat performance in cpu-intensive calculations. However, it's>>> still an order of magnitude slower than Linux when under a heavy load>>> (134 seconds vs 18 seconds).>> We are experience the same result here. WebCat/Linux always runs>> faster the WebCat/Mac OS X Server/Client>> I think it is OS dependent> > OSX is NOT always slower. Running a single instance of my template,> OSX is faster by 50%. It's only when multiple requests occur> simultaneously that OSX crawls, even in comparison to OS9/Webstar on> the same hardware.> > Isn't it possible that some inefficiencies exist in the way> webcat/osx handles threads that could cause this massive difference> in performance? Do other applications that run on both OSX and Linux> have this problem, or is this a problem specific to webcat? If not> in webcat, then the problem should be reported to the Darwin group> and to Apple as a bug.> > If this is an inherent flaw in (current) OSX, I would strongly> discourage anyone from using webcat/osx for busy webcat-driven sites.> It certainly can't keep up with our (relatively simple) testing suite> using Netsplat.> > -Dale-------------------------------------------------------------This message is sent to you because you are subscribed to the mailing list .To unsubscribe, E-mail to: To switch to the DIGEST mode, E-mail to Web Archive of this list is at: http://search.smithmicro.com/ Alain Russell

DOWNLOAD WEBDNA NOW!

Re: macosx 1 process, linux N processes, macosx chokes under load

2001

Top Articles:

Related Readings: