From: | "Luke Lonergan" <LLonergan(at)greenplum(dot)com> |
---|---|
To: | "Joshua D(dot) Drake" <jd(at)commandprompt(dot)com>, "PFC" <lists(at)boutiquenumerique(dot)com> |
Cc: | "Dave Cramer" <pg(at)fastcrypt(dot)com>, "Michael Ben-Nes" <miki(at)canaan(dot)co(dot)il>, pgsql-performance(at)postgresql(dot)org |
Subject: | Re: Advice on RAID card |
Date: | 2005-09-25 18:09:30 |
Message-ID: | 3E37B936B592014B978C4415F90D662D0C4276@MI8NYCMAIL06.Mi8.com |
Views: | Raw Message | Whole Thread | Download mbox | Resend email |
Thread: | |
Lists: | pgsql-performance |
>> Even for RAID5 ? it uses a bit more CPU for the parity calculations.
> I honestly can't speak to RAID 5. I don't (and won't) use it. RAID 5 is
> a little brutal when under
> heavy write load. I use either 1, or 10.
Yes, for RAID5 software RAID is better than HW RAID today - the modern general purpose CPUs are *much* faster at the ECC calculations than the CPUs on most modern hardware SCSI RAID cards.
Note that there is a trend toward SATA RAID, and the newer crop of SATA RAID adapters from companies like 3Ware are starting to be much faster than software RAID with lower CPU consumption.
Use non-RAID SCSI controllers if you want high performance and low CPU consumption with software RAID. The write-combining and TCQ of SCSI is well suited to SW RAID. Note that if you use HW RAID controllers for SW RAID, expect slightly better performance than in their HW RAID mode, but much higher CPU consumption, as they make for terrible JBOD SCSI controllers. This is especially true of the HP smartarray controllers with their Linux drivers.
- Luke
Greenplum
From | Date | Subject | |
---|---|---|---|
Next Message | Michael Stone | 2005-09-25 18:13:52 | Re: Advice on RAID card |
Previous Message | Simon Riggs | 2005-09-25 17:45:30 | Re: [PERFORM] Releasing memory during External sorting? |