Date: Mon, 6 Nov 1995 11:13:26 -0700 (MST) From: Terry Lambert <terry@lambert.org> To: dyson@freefall.freebsd.org (John Dyson) Cc: current@FreeBSD.org Subject: Re: Info on my Sunday commits Message-ID: <199511061813.LAA15483@phaeton.artisoft.com> In-Reply-To: <199511060035.QAA24457@freefall.freebsd.org> from "John Dyson" at Nov 5, 95 04:35:32 pm
next in thread | previous in thread | raw e-mail | index | archive | help
> MNT_ASYNC now means even more on UFS. More of the meta-data is written > asynchronously and the VREG data is now written asynchronously. I think > that even more can be done relatively safely. This can really help on > bulk file restores or perhaps news spool partitions. UFS has a number of places it gratuitously uses sync I/O to fix code errors. One of the engineers here at Artisoft has rolled some changes that result in some significant speed increases with no reduction in reliability and no change to the on disk layout. The changes are pointed out in the Herrin/Finkel VIVAFS work as changes they made to make VIVA faster, but it seems that all but two of the major changes are directly applicable to UFS as well. Check Technical Report Number 225-93 at the University of Kentucky CS Department (ms.uky.edu?). I haven't looked at the MNT_ASYNC changes yet, but based on its previous behaviour, I strongly caution against using MNT_ASYNC except in very special circumstances, like volume copying or backup, etc.. 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?199511061813.LAA15483>