Date: Mon, 19 Apr 2004 17:51:49 -0700 (PDT) From: Roger Marquis <marquis@roble.com> To: freebsd-security@freebsd.org Subject: Re: Is log_in_vain really good or really bad? Message-ID: <20040420005149.AF50FDAFCC@mx7.roble.com> In-Reply-To: <20040420003036.GL18311@cowbert.net> References: <20040417190059.06B0316A4F7@hub.freebsd.org> <20040420003036.GL18311@cowbert.net>
next in thread | previous in thread | raw e-mail | index | archive | help
Peter C. Lai wrote: > > This is hardware problem. Any ATA/SATA disk will suck up CPU with > > every disk access. > > Only if you are running your drive in PIO mode. The system starts up using the > highest UDMA level possible and I bet (hope) he checked the sysctl to make sure > it was at UDMA66. PIO mode, UDMA, 66/100, serial, all are factors but none compensate for the differences beteween SCSI and ATA in real world conditions. By real world I mean where there are multiple, simultaneous disks reads and writes. Too many "benchmarks" only test serial, non-multitasking disk access. In this mode ATA can be just as fast as SCSI and use nearly as little CPU. As soon as you factor in multitasking, however, it's a whole 'nother ballgame. I've seen 10K 160M SCSI drives handle 10x more data than the fastest UDMA100. With 15K drives and 320M becoming generally available SATA isn't even rated for 50% of SCSI's maximum throughput and that's before factoring in multitasking. In every test I've done with various disks and controllers SCSI has out-performed ATA by a wide margin in server performance. -- Roger Marquis Roble Systems Consulting http://www.roble.com/
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20040420005149.AF50FDAFCC>