Date: Mon, 22 Sep 1997 20:45:12 -0600 From: "Justin T. Gibbs" <gibbs@plutotech.com> To: Terry Lambert <tlambert@primenet.com> Cc: gibbs@plutotech.com (Justin T. Gibbs), nate@mt.sri.com, bde@zeta.org.au, current@FreeBSD.ORG Subject: Re: callouts in CAM (was Re: cvs commit:) Message-ID: <199709230245.UAA10200@pluto.plutotech.com> In-Reply-To: Your message of "Tue, 23 Sep 1997 02:16:24 -0000." <199709230216.TAA00163@usr01.primenet.com>
next in thread | previous in thread | raw e-mail | index | archive | help
>> Yes, but it's still not implemented. I think that we should push this >> onto the client instead of attempting to do some kind of low water-mark >> early allocation. In other words, allocate a fairly small initial pool >> for most applications and then have systems like CAM allocate a callout >> on an as-needed basis. > >You don't think it should be watermarked? I am a fan of low watermark >based allocation scheduling (not necessarily immediate allocation, unless >the pool empties). Mostly, I like this because the pools can be per >CPU, and thus you don't take a global resource lock in the SMP case. My point is that clients can allocate or request for allocation deterministically as they know what their usage will be. If there is an interface to do this, then the client can deal with a failure gracefully. If you rely on watermark based allocation and for some reason cannot keep up with demand, there is little you can do other than panic. > Terry Lambert > terry@lambert.org > --- > Any opinions in this posting are my own and not those of my present > or previous employers. -- Justin T. Gibbs =========================================== FreeBSD: Turning PCs into workstations ===========================================
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199709230245.UAA10200>