Date: Thu, 5 Jun 1997 10:00:17 -0700 (MST) From: Terry Lambert <terry@lambert.org> To: gibbs@plutotech.com (Justin T. Gibbs) Cc: ken@housing1.stucen.gatech.edu, freebsd-smp@FreeBSD.ORG, freebsd-scsi@FreeBSD.ORG Subject: Re: 2xPP200 vs 2xPP150 Message-ID: <199706051700.KAA20396@phaeton.artisoft.com> In-Reply-To: <199706051643.KAA28909@pluto.plutotech.com> from "Justin T. Gibbs" at Jun 5, 97 11:41:39 am
next in thread | previous in thread | raw e-mail | index | archive | help
> [About the Quantum Atlas II] [ ... ] > According to Quantum, the Atlas II's firmware should be able to handle up > to 256 tags at a time. Unfortunately, they seem to hold on to the resource > necessary to queue a transaction on writes until they are completely > written to the media even though the drive may return good status long > before this happens. This makes it really tough for the SCSI system to > determine what is an appropriate tag count to use. > > The current algorithm used in CAM is to simply truncate the number of > "openings" when a QUEUE FULL is encountered to the number of transactions > that are still left on the device (e.g. if the 15th transaction caused a > queue full, we'll drop to 14 and not queue any more to the device until we > drop below that number). It may be that, for drives like the Atlas II, > better performance could be achieved by adding some hysteresis on the count > along with a delay in releasing new transactions, but I don't know how > worth while the added complexity would be. >From your description, it seems that it would be worthwhile to seperate hysteresis for read vs. write, if the write problem is localized solely to the write path. Regards, Terry Lambert terry@lambert.org --- Any opinions in this posting are my own and not those of my present or previous employers.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199706051700.KAA20396>