lmb@vsi1.UUCP (Larry Blair ) (02/02/88)
I recently posted this the comp.sys.sun, but I only received one response, so I'm reposting it here in comp.unix.wizards. We recently upgraded the disk controller in our Sun-3 from an Interphase 3200 to a 4200. I'm trying to configure things so that we achieve the tremendous performance improvement that should be possible. When we upgraded, I was able to do an apples to apples that indicates the kind of improvement available (in theory). Using a Fuji 2361A (supereagle) that was formatted with skew=1, I did the following "dd" of a filesystem (length 20Mb) on a completely idle system: time dd if=/dev/rip0a of=/dev/null With the 3200, the elapsed time was 50 seconds. With the 4200 it was 22 seconds. It seems to me that 25 seconds of the improvement was due to the readahead of the 4200. If so, then only 3 seconds of the improvement was from higher transfer rate. It is, of course, true that reading sequential blocks of a raw filesystem, is completely unlike reading a file within a filesystem. My goal is to get the 4200's readahead to actually read the block I'll want next. I think that by changing the "maxconfig" and "rotdelay" settings with "tunefs", I can achieve this. It's not a all clear to me exactly what values I should use. I'm also concerned that I may get the readahead but end up losing a rotation on write. Should I change the skew in the formatting? In addition to the Fuji, we have a CDC FSD-340. I will appreciate all advise and I will attempt to obtain an apples to apples to determine the real improvement. This news group is already overloaded so please, PLEASE, e-mail me your responses. I promise that will repost the helpful answers and I will give you credit for your help. Larry Blair - VICOM Systems Inc. (408)432-8660 2520 Junction Ave. San Jose, CA 95134 {pyramid, oliveb, vodor, mordor, ucbvax}!tolerant!vsi1!lmb