RAID!

[_|_|_]

I'm trying to find info on how Raid 5 works *in the case of 3 data
drives + 1 parity*. I understand how exor'ing 2 data would yield a
parity bit, but the 3data/1parity algorithm eludes me. I can't think
of how you'd have enough info to recover, even if there were such
a thing as a 3-input Exor. <g>

Is this simple? Or is there a website or paper covering it?

 

[Paul Rubin]

I'm trying to find info on how Raid 5 works *in the case of 3 data
drives + 1 parity*. I understand how exor'ing 2 data would yield a
parity bit, but the 3data/1parity algorithm eludes me. I can't think
of how you'd have enough info to recover, even if there were such
a thing as a 3-input Exor. <g>

Yes, there's such a thing as a 3-input xor. xor(x,y,z) = (x xor y) xor z.

If you have

parity = xor(x,y,z)

and you lose the x drive, you reconstruct it the obvious way:

x = xor(parity, y, z).

Similarly for the y and z drives. Is that all you wanted to know?

 

[_|_|_]

Yes, there's such a thing as a 3-input xor. xor(x,y,z) = (x xor y) xor z.

...

Initially, that felt counterintuitive, but you're right, Paul--it's
just an even/odd thing, and you could add as many drives as desired.

Is that all you wanted to know?

No, but while I was typing I realized that the other questions may be
somewhat arbitrary. In the past I've always run Raid 0 with an
'offboard' mirror. IOW, an image on an external server that's backed
up in non-realtime. Safer against viruses/trojans, but drive failures
have bitten a couple times.

I figure Raid 5 should have the same read performance with a good
controller (3ware, probably) but I'm trying to figure out how much
impact Raid 5 will have on write performance. I realize that this is
dependent on the type of operation, etc. and that I'd probably get
numerous answers stating that.

OTOH, if average writes stay a bit faster than writes to a normal
non-Raid drive, then it could still be worthwhile. I was estimating
Raid5's impact on writes as 25 to 50% (giving wide leeway) so it
should still be faster. Open to comments.

LL

 

[Peter]

I figure Raid 5 should have the same read performance with a good

controller (3ware, probably) but I'm trying to figure out how much
impact Raid 5 will have on write performance. I realize that this is
dependent on the type of operation, etc. and that I'd probably get
numerous answers stating that.

OTOH, if average writes stay a bit faster than writes to a normal
non-Raid drive, then it could still be worthwhile. I was estimating
Raid5's impact on writes as 25 to 50% (giving wide leeway) so it
should still be faster. Open to comments.

Which write performance? Sequential (MBps) or random (IOps)?
How many disks?

 

[_|_|_]

I figure Raid 5 should have the same read performance with a good
controller [as Raid 0]
...
but I'm trying to figure out how much
impact Raid 5 will have on write performance. I realize that this is
dependent on the type of operation, etc. and that I'd probably get
numerous answers stating that.

OTOH, if average writes stay a bit faster than writes to a normal
non-Raid drive, then it could still be worthwhile. I was estimating
Raid5's impact on writes as 25 to 50% (giving wide leeway) so it
should still be faster. Open to comments.

Which write performance? Sequential (MBps) or random (IOps)?

I would guess primarily sequential. Video and audio apps. The audio
would have to stream multiple tracks, so there would be seeks when
filling up audio hardware's output buffers. I don't know the size of
those buffers.

How many disks?

Probably 3 + parity. Controllers are 4 port.

 

[Peter]

I would guess primarily sequential. Video and audio apps. The audio

would have to stream multiple tracks, so there would be seeks when
filling up audio hardware's output buffers. I don't know the size of
those buffers.


Probably 3 + parity. Controllers are 4 port.

You may get a decent sequential write speed with GOOD RAID 5
controller.

May see some reviews to get an idea:
http://www.gamepc.com/labs/view_content.asp?id=9500s4lp&page=7
http://www.tweakers.net/reviews/557/23

 

[_|_|_]

You may get a decent sequential write speed with GOOD RAID 5
controller.

May see some reviews to get an idea:
http://www.gamepc.com/labs/view_content.asp?id=9500s4lp&page=7
http://www.tweakers.net/reviews/557/23

Thanks for the links, Peter.

I had seen the first review (gamepc) some time ago. It seemed odd
that they got such weird results on the 3ware 8000 vs 9000. I kept
thinking that the test was flawed, or that data was scewed. If I
recall, they also said that on-mobo RAID would be faster since it
didn't have to go thru the PCI bus. That also seems wrong, given that
onboard RAID is almost invariably software-based. (I don't remember
ever seeing a motherboard with an actual RAID controller on it).
Also, I don't think the PCI bus would be the bottleneck in most cases.

I'll have to research the raid controllers mentioned in the second
article. I'm only vaguely familiar with some of those manufacturers.
I'm surprised to see that most placed higher than 3ware, which I
thought was at or near the top. Are the other controllers really that
much better?

LL

 

[Maxim S. Shatskih]

onboard RAID is almost invariably software-based. (I don't remember

ever seeing a motherboard with an actual RAID controller on it).

Is HighPoint 370 (or newer) software-based?

 

[_|_|_]

Is HighPoint 370 (or newer) software-based?

Are you saying there is a motherboard using the Highpoint?

 

[Maxim S. Shatskih]

Is HighPoint 370 (or newer) software-based?

Are you saying there is a motherboard using the Highpoint?

Epox D3VA. An old VIA-Apollo-133 dual-CPU mobo (like dual PIII-800).

 

[_|_|_]

Epox D3VA. An old VIA-Apollo-133 dual-CPU mobo (like dual PIII-800).

I should get one of those! <g> Actually, a benchmark on TomsHardware
OC'd a Pentium-M (relative of PIII) and it beat the fastest P4s and
AMDs. Intel should take note.

If the chipset on your PIII motherboard was made by Hightpoint, there
may well have been some processing on board. I can't imagine a modern
'consumer' motherboard containing the equivalent of a $400 Raid
processor, though.

Nor can I imagine anyone doing anything but running Raid 1 on a normal
motherboard Raid controller. I suppose there may be some advantage in
Raid 0 if it's necessary to create a 500+gb volume, but it can't be
very fast.

 

[_LL]

After having trouble tracing cables back to a given drive and
determining which drive letter that represents, it occured to me that
someone must have written a utility to do this already.

IOW, show that a Seagate 400gb is attached as master on cable 1 of the
Adaptec ATA controller, etc. Kinda like CPU-Z but for drives.

Anyone know if this exists?

 

[Maxim S. Shatskih]

Windows? Start/Run/devmgmnt.msc/Show Devices By Connection/expand the tree.

UNIXen? Try "atacontrol".

 

[_LL]

Windows? Start/Run/devmgmnt.msc/Show Devices By Connection/expand the tree.

UNIXen? Try "atacontrol".

Brilliant. Thanks, Maxim. Embarrassing that I never noticed that
option in devmgmnt before.

In the course of looking around, I found this program:
http://www.passmark.com/products/diskcheckup.htm
It doesn't do specifically what I wanted, but it may be of use
to some. It does show detailed info on drives. Unfortunately,
it doesn't find devices on other controllers (3ware and Adaptec
PCI controllers, both mapped as SCSI).