Date: Fri, 31 Jan 2003 13:51:52 -0500 (EST) From: Lord Isildur <mrfusion@uranium.vaxpower.org> To: Steve Byan <stephen_byan@maxtor.com> Cc: David Laight <david@l8s.co.uk>, freebsd-fs@FreeBSD.ORG, tech-kern@netbsd.org Subject: Re: DEV_B_SIZE Message-ID: <Pine.3.89.10301311357.A20439-0100000@uranium.vaxpower.org> In-Reply-To: <1BBFD4B2-354C-11D7-B26B-00306548867E@maxtor.com>
next in thread | previous in thread | raw e-mail | index | archive | help
to just get the performance of aligned accesses, we dont need to modify block sizes and such stuff. an an example, read the paper linked to from this; http://www.pdl.cmu.edu/PDL-FTP/stray/traxtent_abs.html (brought to you by the same folks who did soft updates and raidframe) happy hacking, isildur On Fri, 31 Jan 2003, Steve Byan wrote: > > On Friday, January 31, 2003, at 12:59 PM, David Laight wrote: > > > On Fri, Jan 31, 2003 at 11:30:18AM -0500, Steve Byan wrote: > >> There's a notion afoot in IDEMA to enlarge the underlying physical > >> block size of disks to 4096 bytes while keeping a 512-byte logical > >> block size for the interface. Unaligned accesses would involve either > >> a > >> read-modify-write or some proprietary mechanism that provides > >> persistence without the latency cost of a read-modify-write. > > > > There probably ought to be a way of making the larger physical > > size visible to systems that are willing to support larger > > block sizes. That way misaligned transfers would be far less > > likely. > > Yes, of course. But I asked with respect to an issue other than > performance. > > > > One problem to consider is that disks are still partitioned > > on cylinder boundaries. This is largely historic but isn't > > this doen't actually make much sense, since the geometry > > almost certainly varies across the disk and has to be faked > > to fit the ATA CHS limits and (on PCs) the BIOS interface. > > > > However what it does mean is that a partition could easily > > not start on a 8 (512 byte) sector boundary. > > So misaligned transefers are likely even if the filesystem > > itself is using 4k blocks. > > > > On a PC the partitioning will typically have the first one > > starting in sector 63, and the others at multiple of 16065 > > sectors from the start of the disk). > > > > This doesn't bode well for getting any aligned transfer > > at all. > > We understand that problem. It's just a performance issue. My concern > is that even if we handwave the performance issues, there's an > underlying semantic that would not be satisfied if we were to run > existing software, unmodified, on a disk with an underlying 4K sector > size. > > Regards, > -Steve > -------- > Steve Byan <stephen_byan@maxtor.com> > Design Engineer > Maxtor Corp. > MS 1-3/E23 > 333 South Street > Shrewsbury, MA 01545 > (508) 770-3414 > > To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-fs" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?Pine.3.89.10301311357.A20439-0100000>