Date: Wed, 4 Jan 1995 09:52:16 +1100 From: Bruce Evans <bde@zeta.org.au> To: hackers@freebsd.org, terry@uivlsi.csl.uiuc.edu Subject: Re: Topiozone, a new benchmark? Message-ID: <199501032252.JAA14484@godzilla.zeta.org.au>
next in thread | raw e-mail | index | archive | help
> I've heard for a long time that busmastering disk controller cards >reduce the load on a multi-processing OS, but seeing it makes me a believer! >... >ESDI drive, I ran iozone for both the ESDI->12F and the ESDI->MD21->BT445C >configurations. Using the MD21 slightly reduces the disk's size and iozone >score. OK, that's to be expected (the MD21 uses part of the disk for its own >use). But when I ran top during iozone, was I surprised! On my otherwise >idle system running X and "iozone 32m 512", the ESDI->12F configuration left >only about (my rough estimate) 50% idle CPU time (about 30% of the CPU time >was spent in interrupts), while the ESDI->MD21->BT445C configuration left The ESDI Interrupt time would be over 50% if you had a faster drive :-]. I see an overhead of about 50% for 1100K/sec from an IDE drive. Large overheads are to be expected because the maximum IDE transfer rate is low (3.3MB/sec (?)). ISA ethernet cards transferring 1100K/sec have approximately the same overhead for the same reason. EIDE drives have a much larger maximum transfer rate (11MB/sec (?) in PIO mode 3, not much different for EIDE DMA). This should be almost as good as the busmastering SCSI overhead for one or two drives. The BT445S has a transfer speed of 40MB/sec but you would be lucky to find a memory system as fast as that. Try the `systat -vmstat'iozone benchmark. systat shows the cpu ways in a neater way than top. Bruce
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199501032252.JAA14484>