Take the 2-minute tour ×
Server Fault is a question and answer site for professional system and network administrators. It's 100% free, no registration required.

I'm looking at replacing an old file server sharing a single RAID5 volume powered by 6 10K rpm disks with 8MB caches. The replacement will be 2 RAID10 volumes of 4 disks each. The new disks have 64MB caches... but are only 7200 RPM SATA drives (WD RE4). My question, then, is how would you expect the two volumes to perform comparatively?

This is already a done deal, but I ask the question because I'd like to have an idea of what to expect before I deploy this. My gut tells me the new system should be a nice performance upgrade, especially because of the split to two volumes instead of one. I'm also curious how it would compare if I only had one new volume instead of two.

I worry I may be under-estimating the impact of the rotation speed of the old array. I also worry because, while I expect the new system to be faster, it's also much larger, and therefore will eventually see an increase in the absolute amount of data transferred per unit of time... so is it likely to be enough faster to handle this increase?

Since I know it does matter, this is for a traditional file server that will be used for mapped drives. No database, web servers, or anything like that.

If I get the chance, I'll run some actual benchmarks to compare the new and old arrays with real-world usage and post them after deployment, but again, the true test won't be until the new system is handling as much data transfer relative to it's size as the current system.

share|improve this question
2  
The cache will have no significant effect. The cache only needs to be big enough to adapt the transfer rates of the two interfaces and every disk made in the last decade has had a cache big enough to do that. –  David Schwartz Feb 15 '12 at 16:27
    
@DavidSchwartz, +1, I keep trying to explain that to people, but they seem to have blind faith that more disk cache is better. –  psusi Feb 15 '12 at 18:21
add comment

3 Answers

up vote 4 down vote accepted

Why have 2 x 4-disk-R10's - why not 1 x 8-disk-R10 with two logical disks? That'll perform far better and be generally safer too.

Now on to your 10k vs. 7.2k question - while the 7.2k disks will inherently be at least 40% slower on random IOPS (yes not so much slower for sequential work) what I'd be more worried about is that most 7.2k disks have a 30% 'duty cycle' - i.e. they're designed to only be busy 30% of the time and exceeding this level actually has a MASSIVE impact on their MTBF. I can attest to this after seeing literally hundreds of 7.2k disks fail in less than a month on some arrays we had, switching to 10k disks reduced this number down to single figures - so watch out, 7.2 disks aren't supposed to be driven all the time ok.

share|improve this answer
    
It isn't any safer; it still can handle only 1 to 4 disks failing. –  psusi Feb 15 '12 at 18:18
    
Interesting. My intent in using separate arrays was that I'm planning on 1TB disks and still need to support a lot of Windows XP XP only support 2TB drive sizes, so I would have two 2TB arrays. But, on reflection, I could still build this as one big array and divide it in two partitions. –  Joel Coel Feb 15 '12 at 22:13
    
For the duty cycle, I am at least looking at WD's "Raid Edition" drives, and not cheaper desktop-class drives. –  Joel Coel Feb 15 '12 at 22:14
    
I can't change the materials, but after talking it over with some others I can at least do this and put it all in one big array using the same disks. –  Joel Coel Feb 16 '12 at 15:21
add comment

"Performance" is something that heavily depends on workload and is really hard to get a reliable forecast for. File server workloads could be nearly anything - from mostly idle to badly-random-read-write-hammered. But if the users were fine with the old RAID5, they probably would not be unhappy with the RAID10. Although I would not bet on a vast performance improvment - SATA disks have rather high seek times and sustained random load (especially when involving random writes along with random reads) might slow throughput down to a crawl.

Separation of logical drives only makes sense if you want to isolate different kind of workloads (i.e. you do not want workload A to slow down workload B under any circumstances). If you do not need this kind of isolation, performance will benefit from the larger number of spindles in a single logical drive.

BTW: benchmarks are fine for rough orientation, but to really allow for the conclusion that the new disk subsystem performs better than the old one, you would need to average performance data (throughput, number of requests, request times, disk queue length) from the old and the new arrays and compare.

share|improve this answer
add comment

The individual disks probably have the same or higher sequential throughput than the old ones, but slower seek times due to the lower rotational speed. A 6 disk raid5 has about 5 disks worth of sequential throughput, but isn't so good at random IO, especially writes. A 4 disk raid1+0 ( which is really what most vendors mean when they say raid10 ) has the sequential throughput of about two disks, but handles random IO better than raid5.

Given that, I would say that if you are going to use the two raid10s separately, then they will perform quite a bit worse than the old raid5, unless your load is mostly random writes.

share|improve this answer
add comment

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.