On Tue, 1 Jan 2002 21:06, Jeff Waugh wrote: > <quote who="Russell Coker"> > > > I've just done some tests on that with 33G partitions of 46G IDE drives. > > The drives are on different IDE buses, and the CPU is an Athlon 800. > > > > So it seems to me that page size is probably a good buffer size to use. > > Cool! Nothing like Real Proper Testing to prove a point. ;)
;) > I'm surprised the difference between 512b and 4k wasn't greater though; I'm > sure I've had more spectacular differences in the past. I have too. If you use cat then it seems that buffer size has more of an impact (I think I posted to one of the debian lists about the performance benefits of my hacked cat which uses a minimum buffer size of page size). Although that was with an older kernel, and I think that the buffering and caching has been changed since then. If the caching in the kernel works optimally then there should be no difference in wall-clock time between 512b and 4k buffers. My Athlon CPU wasn't really being stressed with 512b buffers (unlike my previous tests with cat - I guess dd is better written). For write caching the write-back cache should bundle everything into 4K writes, the read-caching should be reading ahead (thus sending requests to the disk drive 100K ahead or more) so the results should be the same regardless of buffer size as far as the hardware is concerned. > ... and I won't bring up anything about SCSI or IDE at this point. ;) ;) -- http://www.coker.com.au/bonnie++/ Bonnie++ hard drive benchmark http://www.coker.com.au/postal/ Postal SMTP/POP benchmark http://www.coker.com.au/projects.html Projects I am working on http://www.coker.com.au/~russell/ My home page -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
