Sign in to follow this  
captainmidnight

best raid level for a new workstation

Recommended Posts

I am going to building a new custom workstation based on this Supermicro barebones system:

http://www.supermicro.com/products/system/...YS-7045A-WT.cfm

My system needs to be high performance. My main tasks (related to stock data processing) stress both the CPU (I will be sticking at least one, maybe two, xeon 5420's in this box) as well as mass storage (I routinely process ~100GB spread among, say, 5,000 or so files--so large file sequential reads/writes are what I mainly care about).

Concerning mass storage, it seems that solid state drives are just not cost effective at this point. Plus I am too nervous to trust brand new technologies with my precious data. So I am left with plain old hard drives.

I am thinking of the new single platter 320 GB WD drive (WD3200AAKS)

http://www.wdc.com/en/products/products.asp?DriveID=299

http://www.anandtech.com/storage/showdoc.aspx?i=3236

http://www.newegg.com/Product/Product.aspx...;Tpk=WD3200AAKS

as it is almost as fast as the best SATA drives out there (except maybe for small random file operations; but all hard drives suck for this anyways). Please recommend a drive that you like better than this, however.

I was originally going to buy 2 of these and configure them in RAID0 to get max large file performance.

I do a daily backup of all my data, so this is not as dangerous as it sounds. Nevertheless, if a drive was to fail, it would be a pain to have to reinstall everything. So I am open to considering higher RAID levels, which I need your advice on.

I think that RAID 1 is out because it gives me zero performance improvement. My understanding is that RAID1 not only performs (slightly) worse than a single drive for writes, it also performs no better than a single drive for reads. (This assumes that there is but a single read thread, which will typically be the case for me. However, if you have multiple read threads, such as in typical server applications, then RAID1 can actually have almost as good read performance as RAID0 because any decent controller will concurrently read different files from the individual drives, since they do not data integrity checking on reads.) Is this understanding correct?

In theory, RAID5 looks very interesting. If a 3-disk RAID5 setup could get 90+% of both the read and write performance of a 2-disk RAID0 setup, then I would gladly pay 50% more for the extra drive.

Unfortunately, the web research that I have done one RAID5 is highly confusing. I have come across completely contradictory reports, some claiming RAID5 performance being close to RAID0, others denying it, and others warning about it being highly dependent on your raid controller. Anyone know of some good modern benchmarks (for my box, or at least the controller that I describe below, would be superb).

By the way, the motherboard on that Supermicro box uses a "Intel ESB2 SATA 3.0Gbps Controller". Questions:

--is this actually a "fake hardware raid" controller?

--anyone know how well it is regarded? In particular, is it well suited for RAID5 or pathetic?

--while I will (unfortunately) initially be running windoze xp on it, I ultimately want to run linux (probably ubuntu) on it; will that be a problem? The supermicro link above claims that "RAID 0, 1, 5, 10 support (Windows only)"

If this RAID stuff is ultimately impossible, then I may reluctantly just buy a single velociraptor (which is an utter ripoff at $300; if it was $150 i would not complain).

Share this post


Link to post
Share on other sites
My understanding is that RAID1 not only performs (slightly) worse than a single drive for writes, it also performs no better than a single drive for reads.
True for the former, not necessarily true for the latter if your RAID controller is smart enough to interleave reads on both disks, which most low-end controllers do not.

RAID5 has inherent overhead due to parity calculations and parity writes and is not a high-performance solution. You would be lucky to get single-disk write performance out of a three-disk RAID5 on chipset RAID.

"Intel ESB2 SATA 3.0Gbps Controller". Questions:

--is this actually a "fake hardware raid" controller?

This is a typical chipset-level RAID controller, so yes, it is host-bound and hence is not going to be very fast for RAID5.

If you need maximum disk performance and do not have the $400+ stomach for a real hardware RAID controller (or to run Linux software RAID, which is free but you need to run Linux and a fast CPU :) ) I would grab the new WD Black Series 1TB or the Velociraptor and take the performance hit.

Share this post


Link to post
Share on other sites

Below are some links that are relevant to this discussion:

1) to get good write performance:

"A 'cheat' to get better write performance is to enable the write back function, as opposed to write through. The system writes to the controller memory, the controller issues the write complete signal and deals with calculations and then the actual writes to the array."

http://www.hardforum.com/showthread.php?t=973161

2) this guy measured 3 disk raid5 to be almost as good as 2 disk raid0:

http://www.poromenos.org/node/78

3) a thorough comparison:

http://home.comcast.net/~jpiszcz/20080607/...fied/disks.html

4) warnings about raid5 writes:

"Raid 5 has a very bad write performance if your OS issue non stripe size multiple IO because it has to read the old parity block of your stripe, use it to compute the new parity block, and write the stripe

if you got a large "write back cache backed by battery" on your controller, you may lower the impact of this raid 5 write performance problem by aggregating IO from the cache but, in a multi-user usage, your controller cache can not guess if you are issuing sequential or direct io...and will issue many read-xor-write IO"

http://forums.storagereview.net/index.php?...raid0+benchmark

Share this post


Link to post
Share on other sites
My understanding is that RAID1 not only performs (slightly) worse than a single drive for writes, it also performs no better than a single drive for reads.

True for the former, not necessarily true for the latter if your RAID controller is smart enough to interleave reads on both disks, which most low-end controllers do not.

And would it be a great bet that the Intel ESB2 SATA 3.0Gbps Controller is in the low-end category, with crappy RAID1 performance?

I could not find anything at Intel concerning this controller that directly answers my question; some related links are

http://www.intel.com/support/motherboards/...b/CS-029100.htm

http://www.intel.com/support/motherboards/...b/CS-025445.htm

If you need maximum disk performance and do not have the $400+ stomach for a real hardware RAID controller (or to run Linux software RAID, which is free but you need to run Linux and a fast CPU :) ) I would grab the new WD Black Series 1TB or the Velociraptor and take the performance hit.

What do you mean by "and take the performance hit"? The hit of using a single drive instead of a true RAID solution?

Share this post


Link to post
Share on other sites
What do you mean by "and take the performance hit"? The hit of using a single drive instead of a true RAID solution?
Yes.

Aside from Tech Report's long-dated article on chipset-level RAID I am not sure what else out there would actually confirm the performance (or lackthereof) of the ESB2 (which is an ICH9 derivative, I think?)...

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
Sign in to follow this