Date: Tue, 10 Dec 2002 19:26:27 +0100 (CET) From: Harti Brandt <brandt@fokus.gmd.de> To: Mikhail Teterin <mi+mx@aldan.algebra.com> Cc: Michael.Schuster@sun.com, <freebsd-sparc@FreeBSD.ORG> Subject: Re: why would Sparc be soo sloow? Message-ID: <20021210192414.B5518-100000@beagle.fokus.gmd.de> In-Reply-To: <200212090921.00890.mi%2Bmx@aldan.algebra.com>
next in thread | previous in thread | raw e-mail | index | archive | help
What compiler are you using on Solaris and FreeBSD? cdrecord-Joerg Schilling did a test with error correction RS codes: FreeBSD with gcc is half the speed of Solaris with the Sun compiler on the same machine (the program is computation only) harti On Mon, 9 Dec 2002, Mikhail Teterin wrote: MT>On Monday 09 December 2002 04:06 am, Michael Schuster wrote: MT>= Mikhail, MT>= MT>= (I read -digest, so this may already be answered) MT>= MT>= > I'm puzzled by the poor performance of our SparcIII @900MHz. MT>= MT>= You need to give us some more information to even give an educated guess, MT>= things like MT>= - memory installed MT>= - no# of CPUs installed MT> MT>Two SparcIIIs on the SunFire, one Pentium4 on the Dell. But the program MT>is single threaded/processed. MT> MT>= - other HW config aspects, like storage in use, etc for both (!) MT>= machines, as well as characteristic of the application, such MT>= as MT>= - memory footprint MT>= - I/O behaviour, etc. MT> MT>Very little -- all the program is doing is computations -- the CPU is MT>the bottleneck on both machines. It calls the same financial formula MT>over and over -- for the same set of numbers. The numbers -- 8 double MT>values per call -- are read using scanf(3) from a file, and the result MT>-- 7 double values per call are written with a printf(3) to /dev/null. MT> MT>There are no file being opened/closed by the program at all... MT> MT>= Is anything else happening while you're testing, or do you have MT>= exclusive access to the machines? MT> MT>Yes, there is. But not very much. And, once again, I'm measuring the CPU MT>time, not the total time. MT> MT>= - iostat MT>= - vmstat * MT>= - mpstat * MT>= - prstat MT> MT>= For obvious reasons, I cannot offer to analyse any data for you, I MT>= hope though that this will get you going. MT> MT>I don't believe those utilities are going to help much -- this is a MT>very simple program, with virtually no IO. The simple top(1) shows, MT>that it is only limited by the CPU, and I remain puzzled by the speed MT>difference, as well as by the high "system" time component on Solaris: MT> MT>mi@attila:~/fxpr/test (70) time ./fxprt < ../tasks.txt > /dev/null MT>819.11u 819.41s 28:35.00 95.5% MT>mi@attila:~/fxpr/test (71) /usr/bin/time ./fxprt < ../tasks.txt > /dev/null MT> MT> real 28:27.5 MT> user 13:35.6 MT> sys 13:39.1 MT> MT>Very suprising, for, according even to truss(1), the program makes no MT>system calls at all after the few at startup -- except for the very MT>occasional read(2) and write(2) (to /dev/null). Indeed, on FreeBSD the MT>"sys" part is negligeable. MT> MT>Is there a known problem with sys-time accounting on these systems: MT> MT> mi@attila:~/fxpr/test (76) uname -a MT>SunOS attila 5.8 Generic_108528-13 sun4u sparc SUNW,Sun-Fire-280R MT> mi@attila:~/fxpr/test (77) uname -X MT> System = SunOS MT> Node = attila MT> Release = 5.8 MT> KernelID = Generic_108528-13 MT> Machine = sun4u MT> BusType = <unknown> MT> Serial = <unknown> MT> Users = <unknown> MT> OEM# = 0 MT> Origin# = 1 MT> NumCPU = 2 MT> MT>Yours, MT> MT> -mi MT> MT> MT>To Unsubscribe: send mail to majordomo@FreeBSD.org MT>with "unsubscribe freebsd-sparc" in the body of the message MT> -- harti brandt, http://www.fokus.gmd.de/research/cc/cats/employees/hartmut.brandt/private brandt@fokus.gmd.de, brandt@fokus.fhg.de To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-sparc" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20021210192414.B5518-100000>