On Mon, 12 Jun 2017 04:55:52 -0000 (UTC), [email protected] (Michael van Elst) wrote:
> Currently best practice for such a RAID is: > > -> align the RAID partition(s) on each component to 4k A given. That's all these components are going to do, so 0 % 8 == 0. > -> use a stripe size of 64k (maxphys), with RAID-5 on 3 disks that's > SecPerSU=64 (parity doesn't count). Doesn't RAIDframe assume 512-byte sectors? So, 64k stripe size would be SecPerSU=128, yes? > -> format the filesystem(s) with 64k blocks and 8k fragments (64k fragments > might be faster sometimes but waste disk space). If one is making blocks=stripes, won't a single 8k fragment write provoke an RMW operation at the RF level (operate on less than a full stripe)? True, it won't cause RMW on the disk as it'll operate on full 4k sectors across both data components. I seem to recall another strategy to make frags=stripes so the minimal FS operation would operate on a single stripe and a block would then be multiple stripes. Or does that incur too much parity overhead? That was my idea in my initial query. The idea of SecPerSU=16 being the minimum that would have frags=stripes=8k on a 3-disk RAID-5. Thanks for the feedback. -- |/"\ John D. Baker, KN5UKS NetBSD Darwin/MacOS X |\ / jdbaker[snail]mylinuxisp[flyspeck]com OpenBSD FreeBSD | X No HTML/proprietary data in email. BSD just sits there and works! |/ \ GPGkeyID: D703 4A7E 479F 63F8 D3F4 BD99 9572 8F23 E4AD 1645
