Re: Beating a dead horse

[Thor Lancelot Simon <tls%panix.com@localhost>]

On Thu, Nov 26, 2015 at 08:57:24AM +0000, Michael van Elst wrote:
> wam%hiwaay.net@localhost ("William A. Mahaffey III") writes:
> 
> >Hmmmm .... I thought that the RAID5 would write 1 parity byte & 4 data 
> >bytes in parallel, i.e. no '1 drive bottleneck'.
> 
> That only happens when the "4 data bytes" (actually the whole stripe)
> gets written in one operation.
> 
> Unfortunately this currently requires that the filesystem sends
> a whole stripe to the RAID driver which is only possible for
> a few configurations because of the MAXPHYS limit of 64kbyte.
> 
> The limit, despite its name, also applies to the RAID driver.
> Additionally it helps when the block size (or even the fragment
> size) is as large as the stripe.
> 
> So, with a 4+1 RAID5:
> 
> sectPerSU = 32              -> 4*32 = 128 blocks = 64KByte stripe
> newfs -b 65536 -f 65536     -> format filesystem with 64k blocks and frags
> 
> To overcome this issue either MAXPHYS needs to be larger or
> the raid driver needs to coalesce multiple writes.

Both, I think.  I will try to resuscitate tls-maxphys.

Thor