Subject: Re: UBC problems
To: Chuck Silvers <chuq@chuq.com>
From: Jonathan Stone <jonathan@dsg.stanford.edu>
List: port-alpha
Date: 01/20/2005 11:19:09
In message <20050120155435.GB21669@spathi.chuq.com>,
Chuck Silvers writes:


>the read-ahead code in genfs_getpages is hopelessly primitive
>and should really be redesigned instead of just being tweaked forever.
>I didn't spend much time on it back when I wrote it, and it really shows.

I think one could make a very strong argument that genfs_putpages is
the wrong place to be enforcing such decisions. it's not the right
place to be making decision which rely, ultimately, on the
bandwidth-delay product of the filesystem "pipe".

That bandwidth-delay product can, and does, vary by several orders of
magnitude (think NFS over high-speed long-distance links, vs local
arrays with nonvolatile stable-storage cache) high, and there's not
(AFAIK) any good way to tell the generic filesystem layer about
current estimates of BW*RTT for a given file, or filesystem, or
mountpoint.