Need drives for RAID on server - better than standard desktop SATA, cheaper than SAS

[Kremlar]

Need to configure a server with some decent performing hard drives. Server will have 6 drives (2 in a RAID1 configuration, 3 in a RAID5 configuration, 1 hotspare). I'd like to go a big higher-end than standard desktop SATA drives, but 15K RPM SAS is out of the budget.

I'm considering using this drive and am looking for some opinions:
http://www.newegg.com/Product/Produc...-322-_-Product

My past experience with WD has been iffy, lots of failed drives - but this drive seems to be highly recommended and reportedly very reliable.

Any opionions, or maybe another option to consider?

Thanks!

 

[zuffy]

I have 4 of the 300GB and 2 of the 150GB for the past 2 years. Not one failure. A few of them are running 24/7. Also, it runs cooler and quieter than some of the 7200rpm drives.

 

[Kremlar]

Thanks for the feedback!

 

[Zargon]

you can get the RE3's.

http://www.newegg.com/Product/Produc...82E16822136313

got a few at work, seem to be OK.

 

[pjkenned]

Out of stupidity, why not just Raid 6 + hotspare with 3.5" 1TB+ SATA drives? What controller are you using? Also, remember SAS is parallel, SATA is serial, and modern 15k rpm SAS drives are significantly faster than the Velociraptor. Velociraptors these days, performance wise, are much closer to current-gen large 3.5" 7,200rpm SATA drives than current-gen SAS. That being said, all of my 15k SAS arrays, 2.5" and 3.5" now reside in servers, but I'm very happy with my big 7200.11 raid 6 array. Assuming you are using a LSI/Adaptec/Areca controller that supports raid 6, 3.5" SATA TB+ drives are really fast... and cheap.

 

[Kremlar]

Controller is an Intel-branded LSI SAS/SATA controller. We are going with RAID1/RAID5/hotspare because some of the applications recommend being on 2 separate physical volumes for a few different reasons - partitions wouldn't help. Chassis is limited to 6 drives.

We'd love to go with SAS, but unfortunately it's not in our budget. I realize the cost is not THAT much higher than the WD 10K SATA drives, but unfortunately it's enough of a difference for the decision makers to say no - so I'm trying to get the best performance I can within the budget, while keeping business-class reliability.

Thanks

 

[sxr7171]

If you want get the server grade versions of home drives. But then again I think if you have engineered redundancy then get the cheapest drive possible and replace them when they break.

 

[LokutusofBorg]

They make 10k SAS drives. They are quite a bit cheaper than the 15k ones. Have you looked at these at all?

They also make enterprise-class SATA drives (7200 RPM). If you have to go SATA this is what you should use. WD's RE series, or the RAID class drives from Samsung or Fujitsu. Seagate has some also if you're not afraid of that brand right now (I'm not, just thought I'd qualify that).

Out of stupidity, why not just Raid 6 + hotspare with 3.5" 1TB+ SATA drives? ...

This would be my recommendation also. When talking RAID, a bunch of spindles with logical drives (LUNs, partitions...) will give you much better performance than your RAID1 + RAID5 configuration. That three-drive RAID5 array will crawl, especially if you use standard 7200 RPM SATA drives. Three-drive RAID5 is the antithesis of high performance.

... and modern 15k rpm SAS drives are significantly faster than the Velociraptor. Velociraptors these days, performance wise, are much closer to current-gen large 3.5" 7,200rpm SATA drives than current-gen SAS.

Aye, VRs aren't worth the premium, IMO. Compare the performance of a not-latest-generation VR to the latest generation 7200 drives. I read an article not too long ago (no link, sorry) that did exactly that, and it showed they're almost identical.

... Also, remember SAS is parallel, SATA is serial ...

Huh? SAS = Serial Attached SCSI.

 

[Kremlar]

Thanks so much for the reply.

They make 10k SAS drives. They are quite a bit cheaper than the 15k ones. Have you looked at these at all?

I did, and they are still a bit out of the price range.

They also make enterprise-class SATA drives (7200 RPM). If you have to go SATA this is what you should use. WD's RE series, or the RAID class drives from Samsung or Fujitsu. Seagate has some also if you're not afraid of that brand right now (I'm not, just thought I'd qualify that).

Aren't the VR considered enterprise-class as well? According to WD, they seem to be:

http://www.westerndigital.com/en/products/index.asp?cat=2

This would be my recommendation also. When talking RAID, a bunch of spindles with logical drives (LUNs, partitions...) will give you much better performance than your RAID1 + RAID5 configuration. That three-drive RAID5 array will crawl, especially if you use standard 7200 RPM SATA drives. Three-drive RAID5 is the antithesis of high performance.

I realize 3-drive RAID5 isn't ideal, but I don't want to give up the hotspare and I'm concerned about simultaneous access to 2 partitions. For example, I'd ideally want to split the log files and database in Exchange to 2 different physical arrays. What would be worse? Putting them on separate partitions on a single physlcal RAID6 array, or splitting them between separate RAID1/RAID5 arrays?

While straight disk performance with a speed test app might be better on the RAID6, past experience has proved to me that overall Exchange performance is better with the RAID1/RAID5 configuration.

Or can you convince me otherwise?

Aye, VRs aren't worth the premium, IMO. Compare the performance of a not-latest-generation VR to the latest generation 7200 drives. I read an article not too long ago (no link, sorry) that did exactly that, and it showed they're almost identical.

Past generation doesn't matter to me. Do current generation VR's outperform current generation 7200 RPM SATAs? If so, we'll spend the money because it's in our budget.

Capacity isn't important to us in this situation, as long as we get 900GB or so total disk space at the end of the day with our 6 avaialble drive bays.

 

[zuffy]

This is a good article. It will give you an idea of how the VR compare to the new 7200 rpm drives.

http://www.tomshardware.com/reviews/2tb-hdd-7200,2430.html

IMO, is storage is not as important, I would trust the VR more.

 

[pjkenned]

Huh? SAS = Serial Attached SCSI.

Sorry should say SAS = bi-directional, SATA is uni-directional. Obvious limitation on my brain's multi-tasking abilities. That's why SAS is a higher performance interface than SATA.

 

[MysticLlama]

For Exchange and six drives, I always go with 3 mirrors. The 5 adds unnecessary overhead when only using 3 drives, and then you're sharing the OS/Logs.

I think on some controllers you can make one of the other mirrors, like the secondary OS drive a sort of hot spare where it mirrors the OS unless you lose a log or data drive and then gives protecting that the priority.

Just a couple thoughts.

 

[theevilsharpie]

I realize 3-drive RAID5 isn't ideal, but I don't want to give up the hotspare and I'm concerned about simultaneous access to 2 partitions. For example, I'd ideally want to split the log files and database in Exchange to 2 different physical arrays. What would be worse? Putting them on separate partitions on a single physlcal RAID6 array, or splitting them between separate RAID1/RAID5 arrays?

While straight disk performance with a speed test app might be better on the RAID6, past experience has proved to me that overall Exchange performance is better with the RAID1/RAID5 configuration.

Or can you convince me otherwise?

You will almost always achieve the best performance by combining all of you disks into a single array and then slicing them up with LUNs. This gives the system the ability to use the full performance of your disk system at all times. Entire storage product lines, like the HP StorageWorks EVA, are designed around this concept.

For your particular situation, I wouldn't recommended RAID 5 nor RAID 6, as the write performance is slow, and performance will degrade even more if a disk fails. For applications like Exchange that have an equal amount of random read/write activity, RAID 10 is ideal. Six 500GB 7200RPM disks in a RAID 10 array should meet your storage and performance requirements.

 

[Kremlar]

Ugh, funny how everyone has different opinions.

Thanks guys

 

[SolMiester]

Hmmm, I have to ask how many users you intend on the exchange box?, that requires logs and data on separate arrays?...sql, fair enough, but exchange?

 

[LokutusofBorg]

Aren't the VR considered enterprise-class as well? According to WD, they seem to be:

http://www.westerndigital.com/en/products/index.asp?cat=2

...

Past generation doesn't matter to me. Do current generation VR's outperform current generation 7200 RPM SATAs? If so, we'll spend the money because it's in our budget.

Capacity isn't important to us in this situation, as long as we get 900GB or so total disk space at the end of the day with our 6 avaialble drive bays.

The point I'm making there is that VRs are only marginally faster when the VR is first released, but regular 7200RPM drives quickly catch up, and for a lot less money. You should go look at the graphs in that Tom's Hardware article someone linked.

I realize 3-drive RAID5 isn't ideal, but I don't want to give up the hotspare and I'm concerned about simultaneous access to 2 partitions. For example, I'd ideally want to split the log files and database in Exchange to 2 different physical arrays. What would be worse? Putting them on separate partitions on a single physlcal RAID6 array, or splitting them between separate RAID1/RAID5 arrays?

While straight disk performance with a speed test app might be better on the RAID6, past experience has proved to me that overall Exchange performance is better with the RAID1/RAID5 configuration.

Or can you convince me otherwise?

Ugh, funny how everyone has different opinions.

You've really only gotten two recommendations: RAID6 or RAID10. The 3xRAID1 setup MysticLlama recommended is just a variation on RAID10, and will sacrifice a bit of performance for the freedom gained by that failover spare configuration he describes. You say you're really hung up on having a hot-spare, so you should either go with his 3xRAID1 if your controller will allow you to configure it that way, or RAID6. If you can get past needing a hot-spare then you will see the best performance with a RAID10 involving all 6 drives.

The things you list as concerns about accessing multiple partitions/LUNs on a single RAID array are not very grounded. Stacking up multiple spindles is what RAID is all about. The controllers are built with write caches and other stuff that enhances this aspect of the platform.

 

[Kremlar]

Hmmm, I have to ask how many users you intend on the exchange box?, that requires logs and data on separate arrays?...sql, fair enough, but exchange?

Only 50-60, so performance is not going to be horrendous either way, but I'm trying to maximize performance with the resources available.

You should go look at the graphs in that Tom's Hardware article someone linked.

I did, and in Access Time and I/O performance the VR trounced the other drives. Other benchmarks had their ups and downs. The VRs might not be worth the cost in most cases, but for this I think they are worth it. There will also be a lot of multi-user file access with this server, and I think the VRs will be a good fit.

The things you list as concerns about accessing multiple partitions/LUNs on a single RAID array are not very grounded. Stacking up multiple spindles is what RAID is all about. The controllers are built with write caches and other stuff that enhances this aspect of the platform.

Thanks, and I do appreciate the discussion. I don't think having things like the OS/swap file/transaction logs and databases on separate spindles is "not very grounded". It's best practice for Exchange and pretty much any database I have worked with. To me the benefits are obvious as simultaneous writes/reads from multiple threads to the same array will slow performance considerably. Having those reads/writes divided between spindles should keep performance up.

Perhaps the overall performance gain with a larger RAID5 or RAID6 will compensate for that loss of performance, and that's what I'm trying to figure out. It's pretty much impossible to benchmark because it's not easy to simulate.

Here's a question. If I do go with a single array, why not just a 5 drive RAID5 + hot spare? Or a RAID6 with no hot spare? RAID6 is typically slowere than RAID5 as far as I know becuse of the 2 parity drives, so RAID5 + hot spare should give me better performance. If I want to reduce the risk of failure during an array rebuild of a failed drive, I could go with RAID6. Since there are 2 parity drives in RAID6 I would feel more comfortable without a hotspare.

RAID6+hotspare seems a bit overkill with 6 drives.

 

[taltamir]

would cheaper drives at raid1 (or 10) instead of raid6 be superior performance one and safety wise and be easier to upgrade when needed?

 

[Thor86]

The number of users is important to determine, but also, mailbox size, and usage play critical roles on sizing datastore disks and spindle amounts.

Anything redundant would be fine, but depending on what the missing information from above will be, it would be best you have at least 4 spindles regardless of what type of drive/interface/RAID setup.

Secondly, mechanical disks by nature WILL fail, and is never a question of if, but only when, so irregardless of 15K SAS or 10K SATA or 7200SATA, they will all fail eventually, plan your redundancy, and make sure to have off-storage, site backups.

 

[Kremlar]

Anything redundant would be fine, but depending on what the missing information from above will be, it would be best you have at least 4 spindles regardless of what type of drive/interface/RAID setup.

That's not a question, there will be 6 drives no matter what option I choose.

Secondly, mechanical disks by nature WILL fail, and is never a question of if, but only when, so irregardless of 15K SAS or 10K SATA or 7200SATA, they will all fail eventually, plan your redundancy, and make sure to have off-storage, site backups.

I'm well aware of that. That's not what this thread is about.

 

[Emulex]

raid 1+0 on all 6 drives, plenty of read write iops for that many users. since there are 3 primary drives and 3 mirror drives you could lose up to 3 (of one or the other) and continue to run.

 

[MysticLlama]

Kinda forgot about this for a long time, but I guess I could explain the three volumes a little better.

The reasoning for that is the type of drive access and separating the spindles for that type of access.

Logs on Exchange write sequentially, so if you have your log drives on a separate spindle, you don't have the heads moving all over the place, they are basically always where they are supposed to be.

Your data drives on the other hand are random and bouncing around all of the time. Add logs to this sort of disk behavior and you kill your log write speed, which is what results in noticeable slowness in Exchange since logs are very prioritized.

When I ran Exchange / SQL on a small SAN I did it this way in order to keep the spindles doing what they would in the most efficient way while also gaining the read/write speed from multiple spindles for each:

RAID 10 over 4 disks, 10k, ran Logs for SQL and Exchange as well as backup staging (also linear) RAID 10 doesn't have the write penalties that a 5 or 6 does with parity, and logs are 95%+ writing

RAID 6 over another 7 disks, 15k, ran data for SQL and Exchange, all random use

OS drives were local to the servers and RAID1, page files went here

This kept the price down by not buying all of the fastest drives where not needed, allowing more storage to be purchased.

Of course, getting into it this much is overkill for 50-60 users and any of the above configs offered by everyone will work. I just tried to keep mine as optimal as possible in case of the need to quickly scale.

 

[MysticLlama]

Another thing to mention on Exchange, speed wise:

Spam comprises over half of the mail that comes into most orgs. Deal with the spam either outsourced (Postini, etc.) or on a network device (Cisco ASA + Trend, etc.) before the Exchange server and you'll gain that much performance automatically since it won't have to deal with receiving it, scanning it, and deleting it.