Date: Mon, 29 Aug 2005 15:33:01 -0400 From: Chuck Swiger <cswiger@mac.com> To: Matthias Buelow <mkb@incubus.de> Cc: Don Lewis <truckman@freebsd.org>, freebsd-stable@freebsd.org Subject: Re: Sysinstall automatic filesystem size generation. Message-ID: <431362ED.9030800@mac.com> In-Reply-To: <20050829185933.GB1462@drjekyll.mkbuelow.net> References: <20050829120415.GA1462@drjekyll.mkbuelow.net> <200508291836.j7TIaVEk013147@gw.catspoiler.org> <20050829185933.GB1462@drjekyll.mkbuelow.net>
next in thread | previous in thread | raw e-mail | index | archive | help
Matthias Buelow wrote: > Don Lewis wrote: [ ... ] >> Did you remember to disable write caching by setting the WCE mode page >> bit to zero? At least with SCSI, it doesn't seem to affect performance >> under most workloads. > > No.. I thought that with SCSI it is "ok" to leave the cache enabled > because SCSI supports some sort of request queueing which doesn't > break the order established by softupdates? That gives you the ability to sequence events, yes, but it doesn't mean that data which has been cached by the drive and not yet written out is fine if the power goes away. Good SCSI/RAID controllers have a small battery backup inside the computer case to address exactly this: aac0: <Dell PERC 3/Di> mem 0xf0000000-0xf7ffffff irq 31 at device 2.1 on pci2 aac0: i960RX 100MHz, 118MB cache memory, optional battery present aac0: Kernel 2.5-0, Build 2991, S/N xxxxx That cache will get flushed to disk even if the OS goes bonkers or disappears entirely due to no power. Maybe something like this would make you happier: # cat >> /etc/sysctl.conf kern.filedelay=7 kern.dirdelay=6 kern.metadelay=5 ...? >> Softupdates only tries to guarantee that the on-disk file system is in a >> consistent state at all times, with the possible exception that not all >> space may be accounted for. > > It doesn't try very hard, though, nor is it very successful. Look, there is a tradeoff between price/performance/quality (or reliability) in most circumstances. If you want more reliability, pay more to get good hardware, or accept that there will be performance loss if/when you choose to maximize reliability. It's also true that FreeBSD could do a better job or otherwise be improved. You don't have to argue that point, we're already convinced. Submitting improvements is useful. Would changing the sysctls above to shorter defaults be a good idea? -- -Chuck PS: Haven't we had this conversation before?
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?431362ED.9030800>