[ Wednesday, September 15, 1999 ] CJones wrote:
> In a single user sequential read, you will get no better performance
> that a normal disk, and I would suspect, that with overheads and such
> for the raid device, you will have some degredation. 25% seems like
> more than some though.
Not sure if I can agree here...
In the sequential read case, raid1 becomes raid0 since you can balance
reads across all non-resyncing mirrors... actually, it can *beat* raid0
when tuned since raid1 can't get the read to another drive until we go
outside the current stripe range and that data is actually ON another
drive... in raid1 we can switch however and whenever we want!
My theory now is that the raid code is overly-conservative in trying to
keep disk seek's from happening, and in the pure sequential read case
this hurts us because we don't balance the read load across drives
effectively... in pure sequential read's the seek penalty is *much*
smaller than the typical case, though, since the other disk(s) seek time
to get to the relevant sector is very tiny, since it can only be a block
or two behind in the large sequential read case...
I'm afraid the response will be to keep the seek-avoiding behavior
as-is because while it hurts the bonnie case, it helps a good bit
In Real Life (multi-user, lots seeks, and *any* chance to save them
is a Good Thing)
James "maybe I'm just *really* sleepy" Manning
--
Miscellaneous Engineer --- IBM Netfinity Performance Development
