[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Old Index]
Re: SCSI/RAID cards on a sparc64
-----BEGIN PGP SIGNED MESSAGE-----
On Jan 30, 2010, at 3:50 PM, Chris Ross wrote:
Following slightly behind the conversations about disk and network
performance, I'm realizing that the disk I/O setup in my quad U-II
server is just too low-end.
I have an Ultra E420, with four U-II @ 450 Mhz. Not a very fast
machine, but respectable. However, I'm using the two SCSI drive
bays for a pair of disks that I'm running RAIDframe on. And, the
disk performance is awful.
Better find out why:
- - what kind of disks do you use? U have two 18GB Seagate running as
stripe set in my U60, performance is quite nice.
- - what stripe size do you use? Might need something bigger.
- - do you use softdeps, WAPBL or anything?
I have a i386 Dell PE2650, and I have just extracted half a dozen
(well, 7) sets of a netbsd-5 build (base, comp, games, man, misc,
text, xbase) onto that machine in less than 5 minutes. Where-as the
same zcat|pax operations took nearly 30 minutes on my afore-
My U60 ( 2x 450MHz US-II and 2GB RAM ) extracts a full set of -current
in about a minute, maybe less, with only one CPU being really busy
( not exactly surprising... )
Some of this is surely processing power, but too much of it IMHO is
the raidframe duplication of I/O onto two 40MB/s SCSI disks. Does
anyone have a recommendation for an U-160 card that I can put into
my E420, or better-yet a NetBSD-configurable/control-able/[monitor-
Watch disk and raid throughput in systat vmstat - I seriously doubt
you're getting anywhere near using up 40MB/s if performance is as bad
as your story indicates - an U160 controller would buy you nothing and
probably wouldn't be bootable.
RAID card that I could use. There's already a two-disk SCA cage in
the machine, so SCSI would be easier from a hardware point of view,
but other options could be considered, especially for cost reasons.
It's always good practice to check your cables, connectors and
terminators. Either one of them is quite likely to mess with
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.7 (Darwin)
-----END PGP SIGNATURE-----
Main Index |
Thread Index |