Date: Thu, 15 Apr 1999 10:01:30 -0700 From: Kent Stewart <kstewart@3-cities.com> To: drkhoe@gmsnet.com Cc: Michael Slater <mikey@iexpress.net.au>, "'freebsd-questions@freebsd.org'" <freebsd-questions@FreeBSD.ORG> Subject: Re: Fact or Fiction (Unix vs NT) Message-ID: <37161B6A.BCA9C1CA@3-cities.com> References: <199904150654.XAA23108@gms.gmsnet.com>
next in thread | previous in thread | raw e-mail | index | archive | help
"Dr. Mosh" wrote: > > >> > LOS GATOS, Calif., April 13. Today, Mindcraft released the results > >> > of a study comparing the performance of Red Hat Linux 5.2 (updated > >> > to the Linux 2.2.2 kernel) and Microsoft Windows NT Server 4.0 > >> > operating systems. According to the report, Windows NT Server > >> > provides over three and a half times the performance of Linux as a > >> > Web server. Furthermore, the report shows that when testing Windows > >> > NT Server and Linux as file servers, Windows NT Server provides over > >> > two and a half times the performance of Linux. The full report, > >> > including all of the details needed to reproduce the tests, is on > >> > Mindcraft's Web site at: > >> > > >> > http://www.mindcraft.com/whitepapers/nts4rhlinux.html. > >> > > > What I find interesting about this whole article is that it was "sponsored" > by Microsoft, and if you look under their Apache configuration, > MaxSpareServers = 290 > > That means it would fire up potentially 290 spare threads for each > request, in effect throttling Linux's kernel... > > This puts their whole Linux/Unix know how in doubt, also, they claim Linux > only used 960megs of the 4gigs of RAM, when a kernel recompile could've fixed > the problem. I doubt if they understood the effective use of swap space > either... > The 960MB doesn't matter because they limited NT to using only 1024MB and 64MB isn't a significant difference. The threads was an obvious problem from the graphs when they went above a midrange number. I have done a lot of benchmarking on a Cray XM/P. You don't know what the important knobs are until you try them all. On our Cray, for example, 50% of the system throughput was dependant on write behind disk caching. You have to provide the running program it's input data and not force writing the output data to disk. The 20GB Cray DD40 disk systems, at that time, were manufacturered using 16 CDC Hyrdra's, which were stripped in groups of 4 to produce 20MB/sec continuous disk I/O. Cray data channels were 1000MB/s and 100MB/s. At the time of the benchmarks ~1988 a Hydra cost $250K US. Cray later replaced the 16 8" full height drives with these little 5.25" half heights that cost on the order of $1000 US while maintaining throughput. Maintanence on the Hydra's was more than the new drives cost. You can imagine what the little HD's looked like in the big 8" bay. You can also imagine the drop in the amount of heat generated by the DD40. One of the reasons I'm using FreeBSD is because I was told by many people that Linux is unstable under heavy loads and that FreeBSD isn't. Whether that is true or not, it is what many Unix people believe. My ISP went from Linux to BSDI and their system stability went up a factor of 10 from my point of view. I was trying to come up with a better solution for a friend that hires me occasionally to do consulting. I thought FreeBSD would make an ideal platform for his people where everyone could telnet in and schedule their jobs. He doesn't have terminal server running on his version of NT. Many of the problems run for a day or more. They have also been known to tell a job to run all weekend. What I have found so far with my comparisons of FreeBSD and NT is that my FORTRAN compiles are running on the order of 15:1. I have one program (the most used one of course) consisting of 299 modules that requires 1.25 hrs for a complete build on FreeBSD and 4-6 minutes with DEC Visual Fortran on NT. FreeBSD is running on a P-166 with 96MB of memory and my NT server has 2-133's with 128MB. The NT server is really light on memory for this kind of activity. I haven't tried building and running the sample problems on a single cpu P-II 400 workstation where the server processes aren't getting in the way. It also wouldn't be fair because I don't have a 400 that I am willing to install FreeBSD on right now. The difference in build times that I am seeing is consistent with make processes where the compiler's are fired up for each module. The ratio can be much worse. The GNU f77 compiler supplied with FreeBSD really isn't a compiler but a frontend to f2c and gcc. Loading and unloading f2c and gcc consumes much time and many resources. You may feed f77 a string of modules to process but the way processing is done requires individual loads. Executions speeds are not favorable either but a factor of two like I am seeing is usually a bad optimization option. Since I am using defaults in both cases, both compiler's likely have a faster combination. The factor of two is about the same range as the Intel Pentium/iMac benchmark comparison. Apple was making a big deal about the speed of their iMac and the Intel Pentium. It turned out that Apple was compiling the PC program for a 386, which was the default cpu, and when PC Magazine used the 586 instruction set, the PC benchmark was suddenly 4 times faster. The Intel machine went from half the speed of the iMac to twice the speed. This also indicates that they were probably much closer overall because each would have it's + and -'s and benchmark differences have a tendancy to be smoothed out in the real world. State changes on the Motorola cpu's have always been better than Intel. Kent > -The Doc > > -- > ------ drkhoe@gmsnet.com -------------- ++++++ ---------------------- > ///// http://progmetal.gmsnet.com ----------------==== Unix systems - > C/C++ video game engine development =><=============== Administration > ===================== Intranet/Internet Engineering ================= -- Kent Stewart Richland, WA mailto:kstewart@3-cities.com http://www.3-cities.com/~kstewart/index.html Hunting Archibald Stewart, b 1802 in Ballymena, Antrim Co., NIR http://www.3-cities.com/~kstewart/genealogy/archibald_stewart.html To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-questions" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?37161B6A.BCA9C1CA>