Date: Sat, 4 May 1996 22:20:57 -0700 (PDT) From: "Rodney W. Grimes" <rgrimes@GndRsh.aac.dev.com> To: hsu@clinet.fi (Heikki Suonsivu) Cc: hdalog@zipnet.net, freebsd-scsi@freebsd.org Subject: Re: Buffered Writes? Message-ID: <199605050520.WAA01785@GndRsh.aac.dev.com> In-Reply-To: <199605050055.DAA13322@cantina.clinet.fi> from Heikki Suonsivu at "May 5, 96 03:55:53 am"
next in thread | previous in thread | raw e-mail | index | archive | help
... > of power failure (unless your drive can write during spindown). > > queue algorithm modifier usually is "write in issue order" so just setting > WCE to 1 does not do cause any damage (and nor it usually improves the > performance if the load is continous like news disk). But the more > important issue is that I still haven't seen a disk which actually *does* > have a write cache (pointers welcome :). Quantum drives are one of the few that both ship from the factory with WCE set to 1 (since there logic and spin down protection circutry actual work). Your following test should work and yeild patter speed/2 as long as you stay on track when using a Quantum Empire, Capella, Saturn, Atlas or Atlas II series drive (it may apply to other Quantum models as well). Most other vendors drives either don't have a true write cache, or they ship the drive with both WCE turned off and the memory (often called the buffer) dedicated to read operations only. The mode pages to get them to write cache are vendor specific and without the vendors manual for the drive your going to have a very hard time to get it turned on. > Easy test is to write data to > every second block on the disk. If there is a write cache, you should get > more than block / ~70ms (one disk revolution) written, optimally you should Pretty darn slow disk drive there, I think you meant 16mS (3600RPM) or 11.1mS (5400RPM) or perhaps 8mS (7200RPM). > get the disk platter speed / 2, but scsi handshake overhead usually makes > this impossible. SCSI handshake overhead is in the noise when looking at modern scsi disk drives and controllers. Depending on how fast you release the I/O's to the drive you can see a whole track written in 1 revolution of the disk when writting all the even sectors followed by all the odd sectors. To do this test correctly you must take into account the ZBR nature of the drive and the cache in the drive must be at least 1 track in size (they usually are closer to 1 cylinder in size total, and depending on how much is segmented for write purposes, etc.). -- Rod Grimes rgrimes@gndrsh.aac.dev.com Accurate Automation Company Reliable computers for FreeBSD
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199605050520.WAA01785>