Date: Tue, 15 Jun 1999 00:12:51 -0700 From: Mike Smith <mike@smith.net.au> To: hackers@freebsd.org Subject: ZD labs test update Message-ID: <199906150712.AAA00708@dingo.cdrom.com>
next in thread | raw e-mail | index | archive | help
Straight back from Usenix, I've returned to ZD labs to continue with the benchmarking attempt we started the week before last. Just to remind everyone, this is the standard Samba-and-Apache runaround with the addition of Zeus to the mix in order to get a feel for its relative performance. The goal here is to add a FreeBSD column to the tests published in PC Week a little while back. The system configuration is as follows; - Compaq Proliant 6100 (an engineering sample) with 4x500MHz PII Xeons and 2GB of RAM. - One Seagate Cheetah for the system disk, on an Adaptec 2940U2W. - Seven Seagate Cheetahs for the data volume, managed by an Infortrend 3201U2, organised as a single RAID5 volume (this is a test requirement) and hung off another 2940U2W. - Four Intel EtherExpress Pro/100's. The RAID controller was kindly loaned by Telenet Systems, the rest of the equipment is provided by ZD labs themselves. Today's progress was greatly aided by Peter Wemm, who managed to miss his flight out last night, and was thus dragged along. This gave us a fighting chance against the Compaq system, which fought us at almost every turn. I'm sure these systems are wonderful servers, but they've got to be towards the "more painful" end of the spectrum when it comes to setting them up. We'd previously encountered problems with the Infortrend controller not at all liking the other disks we'd tried to talk to; a collection of Cheetahs with IBM and Compaq firmware simply wouldn't work. This time we had better luck with real Seagate firmware, and the array performance wasn't too shabby, giving us ~8MB/sec write performance and ~16MB/sec read performance using the dd-stone benchmark. Following established lore, it was necessary to use the secret Ctrl-A hotkey to enable Advanced mode in the Compaq setup program so that the MP table could be set to 'Full Table' mode, as well as rearrange interrupts so that nothing useful was given IRQ 9. This latter was necessary as otherwise the system was presented with around 100,000 interrupts per second, from an unknown source. This consumed about 6% CPU (according to systat) in SMP mode, and caused the peripheral with that IRQ to fail miserably. A few other minor tweaks were performed (upping the buffer cache sizing by setting NBUF to 32,000) and a set of kernels built for UP, 2-way and 4-way SMP. Thus, the ground is prepared for us to begin testing with Samba tomorrow. More news as it comes... -- \\ The mind's the standard \\ Mike Smith \\ of the man. \\ msmith@freebsd.org \\ -- Joseph Merrick \\ msmith@cdrom.com To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-hackers" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199906150712.AAA00708>