Raid 1 or Raid 5

S

Steve

Ok... I've read all the advantages and disadvantages of Raid 1 and
Raid 5. I have (3) 250 GB Serial ATA Drives and an Adaptec 2410SA Raid
controller. I'm putting an array into a Dual Xeon 2.8 Ghz Server using
an Intel Mainboard.

My applications are mixed. We have a law firm. Most of our users work
in MS Word and in an application called AMICUS. It keeps the
attorney's calendars, contacts, notes etc. This database resides on
the server. We have 4 lawyers each with their own assistants. The
other application is a document archival application. Basically,
anything that goes in or out of the office that is paper based is
scanned into the system and ends up in 'cabinets' on the server. The
application is called IMAGEWARE by Canon. Each lawyer has his/her own
cabinet which contains his/her information.

With that said, I'm debating using Raid 1 because it doesn't have the
downside of slower writes like Raid 5 which is doing parity writes. If
I use Raid 1, I'd place 2 drives into the array and the third 250 GB
drive I could configure as a Hot Spare which would automatically kick
in if needed due to a failure on drive 1 or drive 2. I'm not overly
concerned about having more than 250 GB available (RAID 5 with all
three drives would afford me 465GB or so).

I like the idea of stripping however. I'm not sure if in the real
world the increased read performance would offset the writing of
parity structures? Does anyone have a feel for this ? Would the RAID 5
actually perform better than the Raid 1 in my environment ?

With regards to a failure would the RAID 1 scenario with hot spare be
easier to recover from ?

Finally - should I used Write Caching (this pertains to either
scenario). When I went to create my array, adaptec pops up a somewhat
intimidating message about data integrity problems in the event of
power failure. I'll have a battery backup UPS attached to server but
is the performance gained by leaving this in the write enabled
position worth the downside risk ?

Lots of questions.... Thanks in advance for any suggestions !
 
A

Arno Wagner

Previously Steve said:
Ok... I've read all the advantages and disadvantages of Raid 1 and
Raid 5. I have (3) 250 GB Serial ATA Drives and an Adaptec 2410SA Raid
controller. I'm putting an array into a Dual Xeon 2.8 Ghz Server using
an Intel Mainboard.
My applications are mixed. We have a law firm. Most of our users work
in MS Word and in an application called AMICUS. It keeps the
attorney's calendars, contacts, notes etc. This database resides on
the server. We have 4 lawyers each with their own assistants. The
other application is a document archival application. Basically,
anything that goes in or out of the office that is paper based is
scanned into the system and ends up in 'cabinets' on the server. The
application is called IMAGEWARE by Canon. Each lawyer has his/her own
cabinet which contains his/her information.
With that said, I'm debating using Raid 1 because it doesn't have the
downside of slower writes like Raid 5 which is doing parity writes. If
I use Raid 1, I'd place 2 drives into the array and the third 250 GB
drive I could configure as a Hot Spare which would automatically kick
in if needed due to a failure on drive 1 or drive 2. I'm not overly
concerned about having more than 250 GB available (RAID 5 with all
three drives would afford me 465GB or so).
I like the idea of stripping however. I'm not sure if in the real
world the increased read performance would offset the writing of
parity structures? Does anyone have a feel for this ? Would the RAID 5
actually perform better than the Raid 1 in my environment ?

Not necessarily. You can do striping reads from RAID1, since you have
the same data on two disks. Depends on whether the controller does it
or not.
With regards to a failure would the RAID 1 scenario with hot spare be
easier to recover from ?

What do you mean by "recover"? RAID1 + hot spare can tolerate
2 dead drives out of 3, if they do not fail reight after each other.
RAID5 can tolerate one dead drive.
Finally - should I used Write Caching (this pertains to either
scenario). When I went to create my array, adaptec pops up a somewhat
intimidating message about data integrity problems in the event of
power failure. I'll have a battery backup UPS attached to server but
is the performance gained by leaving this in the write enabled
position worth the downside risk ?

You have write caching anyway. Unless you disable it in the disks
themselves. The loss in performance is dramatic.
Lots of questions.... Thanks in advance for any suggestions !

Personally I like RAID1 with hot spare. I do this on some critical
servers. The advantage is that you have time to repair it. With
regard to speed, I would expect both set-ups tp be about the same
speed on reads and writes. The RAID5 has to write 1.5 times the data,
but has 1.5 times the disk bandwidth, so that evens out. For reading
the RAID5 could theoretically be a little faster, but practically I
would expect not much difference.

One note: You need a spare controller! If a hardware RAID controller
fails, you may need an identical controller, or at least one from the
same vendor to access your data. Quite a surprise to many people...

Arno
 
S

Steve

You have write caching anyway. Unless you disable it in the disks
themselves. The loss in performance is dramatic.

I am able to disable disk write caching. Would it get ugly though with
regard to performance.

Since my original post I found a recent review on X-Bit Labs site. It
was quite detailed and admittedly I got a bit lost. They concluded
that the Raid 1 level performance on this particular controller was a
bit puzzling. It performed much like a single drive. There did not
seem to be any optimized algorithm for reads in the Raid 1 scenario
but oddly Raid 10 showed gains in this area which to them indicated
that maybe in a future bios release the Raid 1 would catch up. Should
I still consider Raid 1 with this controller - does this change your
thinking or is it different from what your experience with Raid 1 on
other controllers might be. Maybe check out the article if you have a
chance and let me know what you think. Thanks in advance.

http://www.xbitlabs.com/articles/storage/display/adaptec-2410sa_15.html
 
T

Trazgor

Steve said:
I am able to disable disk write caching. Would it get ugly though with
regard to performance.

Since my original post I found a recent review on X-Bit Labs site. It
was quite detailed and admittedly I got a bit lost. They concluded
that the Raid 1 level performance on this particular controller was a
bit puzzling. It performed much like a single drive. There did not
seem to be any optimized algorithm for reads in the Raid 1 scenario
but oddly Raid 10 showed gains in this area which to them indicated
that maybe in a future bios release the Raid 1 would catch up. Should
I still consider Raid 1 with this controller - does this change your
thinking or is it different from what your experience with Raid 1 on
other controllers might be. Maybe check out the article if you have a
chance and let me know what you think. Thanks in advance.

http://www.xbitlabs.com/articles/storage/display/adaptec-2410sa_15.html

I was told that the 3Ware controllers are far more performant than Adpatec's
ones.
In Europe they are about the same price (750x, 850x series)
 
A

Arno Wagner

I am able to disable disk write caching. Would it get ugly though with
regard to performance.

Yes, unless you have write caching in the OS (there it is again...).
Seriously, write caching is usually not a problem at all. With some
journalling systems write-reodering can cause trouble. If
you use a journaling file system you may have the option of using
a small journal device with fully synchronous writes and a large system
with whatever caching, buffering and reordering.
Since my original post I found a recent review on X-Bit Labs site. It
was quite detailed and admittedly I got a bit lost. They concluded
that the Raid 1 level performance on this particular controller was a
bit puzzling. It performed much like a single drive. There did not
seem to be any optimized algorithm for reads in the Raid 1 scenario
but oddly Raid 10 showed gains in this area which to them indicated
that maybe in a future bios release the Raid 1 would catch up. Should
I still consider Raid 1 with this controller - does this change your
thinking or is it different from what your experience with Raid 1 on
other controllers might be. Maybe check out the article if you have a
chance and let me know what you think. Thanks in advance.

Sorry, I have no experience with hardware RAID. I can tell you
however that RAID1 is also totally unoptimized in Linux software-RAID1
(or at least seem to be). Software RAID5 is uo to n-1 times faster on
reading, and usually slower on writing than a single disk.

Arno
 

Ask a Question

Want to reply to this thread or ask your own question?

You'll need to choose a username for the site, which only take a couple of moments. After that, you can post your question and our members will help you out.

Ask a Question

Top