RAID 5 (yes, I know)

[Laszlo G. Szijarto]

I've been reading Google on Informix raw storage over RAID 5 and keep
encountering the mantra "NO RAID 5". But I can't make out why. I've seen
references to poor performance on (software) RAID 5 as well as discussions
about the threat of data loss if 2 disks go bad.

I have a hardware RAID controller and
data replication between two servers

Now, running over RedHat Linux 6.2, I want to set up a hardware RAID 5 over
the top of which runs one partition for formatted files along with another
unformatted partition on which I would like to write Informix raw data.

Should I not do this? Why?

Thank you for any input you may have,
Laszlo

[JonMEsq]

I have set up RAID 5 on 6.1 using drivers from the raid chip manufacturer
(DPT)-- it seems to work with informix sql and se with no degradation that i
can note. I have not had any trouble (so far) now that it is set up. Setting
it up was a bit of a bear. However, I have never needed it. At NASA, you
probably need r.t. backup. Here, the people were oversold on the hardware, and
a tape or CD-RW would have done it. - Jon
jon...@csd.net, jon...@aol.com

[Art S. Kagel]

Here's a repost of my quarterly NO RAID5 RANT, see attached.

Art S. Kagel

[Glenn Brackenridge]

As ever, some really good observations, but your info on RAID3 and 4
is wrong. RAID4 does not read all the disks for any block - so the same
deficiency exists as with RAID5. With regard to RAID3, there are only
a couple of pure RAID3 mainstream vendors (as opposed to RAID5 vendors
who have a RAID3 option - basically loading cache on the top of a RAID5)
and only one of them actually bothers to do the parity check.

Techniques in RAID5 to defeat the integrity issues are well researched.
These include background scavenging and additional stripe integrity
generation codes typically taking the disk sector to 520 bytes. Some
RAID vendors feel comfortable to offer 100% guarantees.

GB

"Art S. Kagel" <ka...@bloomberg.net> wrote in message
news:3C5733AB...@bloomberg.net...

>
> RAID3 and RAID4 are superior in both areas. In both all drives are read
> for any block which improves sequential read performance (Informix Read
> Ahead depends on sequential read performance) over RAID5 and parity can
> be (and in most implementations IS) checked at read time so that partial
> media failure problems can be detected. Write performance is
> approximately the same as RAID0 for large writes or smaller stripe block
> sizes.
>
> Art S. Kagel

--
Posted via Mailgate.ORG Server - http://www.Mailgate.ORG

[Art S. Kagel]

Glenn Brackenridge wrote:

> As ever, some really good observations, but your info on RAID3 and 4
> is wrong. RAID4 does not read all the disks for any block - so the same
> deficiency exists as with RAID5. With regard to RAID3, there are only
> a couple of pure RAID3 mainstream vendors (as opposed to RAID5 vendors
> who have a RAID3 option - basically loading cache on the top of a RAID5)
> and only one of them actually bothers to do the parity check.
>
> Techniques in RAID5 to defeat the integrity issues are well researched.
> These include background scavenging and additional stripe integrity
> generation codes typically taking the disk sector to 520 bytes. Some
> RAID vendors feel comfortable to offer 100% guarantees.

And having lost many nights sleep using arrays by one of these,
indeed using 520 byte blocks, until we switched to RAID10, I am
even more confident in saying: It does not work reliably enough
for me! And that's not counting the 50% write penalty and the
80% recovery performance penalty!

Art S. Kagel