Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP Path: utzoo!mnetor!seismo!columbia!cheshire.columbia.edu!yoram From: yoram@cheshire.columbia.edu (Yoram Eisenstadter) Newsgroups: comp.arch Subject: Re: Kernel build times, esp. on Amdahl Message-ID: <4794@columbia.UUCP> Date: Tue, 7-Jul-87 22:22:39 EDT Article-I.D.: columbia.4794 Posted: Tue Jul 7 22:22:39 1987 Date-Received: Fri, 10-Jul-87 07:20:52 EDT References: <502@winchester.UUCP> Sender: nobody@columbia.UUCP Reply-To: yoram@cheshire.columbia.edu (Yoram Eisenstadter) Followup-To: comp.arch Distribution: na Organization: Columbia University CS Department Lines: 44 Keywords: benchmarks Summary: Amdahl's numbers seem quite reasonable. In article <502@winchester.UUCP> mash@mips.UUCP (John Mashey) writes: >I've been intrigued by Amdahl's recent recruiting ads that say they >build a UNIX kernel in 3 minutes. Can anybody from Amdahl say: >1) what model this is on? >2) is 3 minutes CPU time, or real time, and what's the other number? >3) is this optimized or unoptimized? > >For comparison: >UNIX kernel make from scratch times: 4.3BSD+NFS, MIPS M/1000: >239.7u 74.5s 12:39 41% 99+231k 6739+6986io 3167pf unoptimized >492.6u 100.3s 18:53 52% 174+313k 8591+8880io 3901pf optimized -O2 I'm not from Amdahl, but here goes anyway... Since C compiles do a great deal of I/O (reading in source and .h files, writing out temp files, reading in temp files, writing out object files) it seems to me that any benchmark of a kernel-make would be heavily dependent on the kind of disk drives attached to the processor. If Amdahl's figures come from a system that's similar to current large IBM mainframes (I don't know much about Amdahl's products, except that they're 370-architecture compatible), they would typically have fast disk drives with large, expensive controllers hooked up via fast block-multiplexor channels. I doubt that MIPS boxes come with such I/O devices. (Amdahl may even have CPUs which are faster than MIPS's; who knows? :-) ) By the way, note that John's benchmarks show that the CPU is idle about half the time, probably waiting for disk reads to complete. Also, assuming Amdahl's numbers represent wall-clock time (and their kernel is as big as 4.3BSD+NFS), they are only 4 to 6 times faster than John's numbers. The fact that the largest conventional mainframe is 4 to 6 times faster than a small RISC processor (is the MIPS a single-chip CPU? single board?) doesn't really surprise me very much. Cheers..Yoram Yoram Eisenstadter | Arpanet: yoram@cs.columbia.edu Columbia University | Usenet: seismo!columbia!cs!yoram Dept. of Computer Science | Bitnet: yoram%cs.columbia.edu@WISCVM New York, NY 10027 | Phone: (212) 280-8180