Matched or unmatched drives for RAID arrays?

19

1

Looking around there is conflicting information on this, with some strongly suggesting one or the other.

From my understanding the issue with matched drives is that the wear on both drives is more or less the same, so the potential for the second drive failing with or very soon after the first is pretty high.

People also claim matched drives give substantially higher performance however assuming the unmatched drives are more or less the same (e.g. 2, 1 TB SATA II 7200rpm drives with 32MB cache), would the minor differences between say a Seagate and a Western Digital one (say one has a 128MB/s read rate, and the other a 150MB/s read rate, as well as I guess various other minor differences) actually cause any notable performance loss, i.e. potentially worse than two matched 128MB/s drives, or does RAID not really care and give you essentially an optimal solution (e.g. up to 278MB/s total read speed for RAID 0 and 1) and similar for other RAID with more "unmatched" drives (5 and 1+0 come to mind as possibilities)?

Also I couldn't find much info on how this is different on different RAID setups, e.g. RAID 0 or RAID 1, software or hardware RAID, etc. I'm assuming such things have an effect, and that's it's not all the same for RAID in general?

Will

Posted 2011-03-15T15:00:16.657

Reputation: 191

You can buy the same model one new and one used for raid1 if that's a problem. – inf3rno – 2017-08-09T18:39:04.320

Answers

17

Points to remember for a RAID 1:

The SLOWER of the drives... will dictate the overall speed.

The SMALLER of the drives... will dictate the overall size.

If you are ok with that.... it's far better TO have drive as different as possible. Purchased at different times, from different stores, made by different manufacturers.

Reason? When the drives fail... I WANT them to fail at different times... not at the same time.

Super_PC_Chick

Posted 2011-03-15T15:00:16.657

Reputation: 171

What about read? I assume it depends of the controller/driver, but if read data from the two disk is NOT checked/compared by the controller, read speed could be equal to the fastest disk, right? And if you have write-cache enable (even in OS), write speed might not suffer that much if one drive is slow. If this is true, one could mix a cheap low speed + expensive faster long-durance drive, and have a good and cheap performance setup. But I again, I guess it depends on the controller/driver. It could be interesting to make a test-setup. One could start with the Intel onboard sata contr. – MrCalvin – 2017-02-11T12:47:35.600

Not true for btrfs software RAID. – inf3rno – 2017-08-10T00:18:00.493

@inf3rno Why that? – dim lost faith in SE – 2019-12-05T12:21:49.117

@dim It can use disks having different sizes and use all the space on them if you have enough disks. – inf3rno – 2019-12-05T16:19:37.510

@inf3rno right, but it does not invalidate the suggestion to get disks from different brands/batches. I thought you were talking about this. – dim lost faith in SE – 2019-12-05T16:52:12.500

@dim No, it does not invalidate it. – inf3rno – 2019-12-05T17:01:38.950

3

The advantages of matched drives are more of a value-economic stance. The array is only going to be as fast as the slowest drive. Other factors are logical, not physical (misaligned stripe array).

surfasb

Posted 2011-03-15T15:00:16.657

Reputation: 21 453

@ZekeHansell How come RAID will be slower? Shouldn't RAID 1 give faster read speeds compared to a single drive? – Pavel Tankov – 2016-08-26T14:18:30.540

So given say a 200MB/s drive and a 150MB/s drive your saying in RAID 1 for example the maximun read speed would be 300MB/s, not 350MB/s? – Will – 2011-03-15T15:06:25.140

The reads and writes have to happen more-or-less simultaneously to the drives. Certain controllers could in theory (and possibly do in practice) handle this at the controller level, but if one of the drives is slower you'd just end up waiting on it anyway. – Shinrai – 2011-03-15T15:12:10.977

2All of this is higly dependent on the raid controller in use. Some raid controllers have a lot of cache, so they can lessen the effect of waiting on the drive, but something to remember is that because it is raid, it WILL BE SLOWER than direct access to the drive EVEN IF the drives are matched in speed. The raid controller becomes a big piece of the performance puzzle. – Zeke Hansell – 2011-03-15T15:16:03.197

Zeke brings up a good point. Depending on the RAID Controller, the differences in unmatched drives can be heavily mitigated. – surfasb – 2011-03-15T21:43:03.580

2

Using matched drives is not important. Especially with "normal conditions"

  1. Value-economic stance: unless you are using really high-end devices (let's say SSD) with low-end drives (old hard disk), you are not saving by using matched drives, at least not substantial amount. Taking 1TB drive with 32MB cache from two manufacturers tend to be about equally priced.

  2. Failing: it's possible disks from same batch break down at the same time. Probability is higher, but not really high. For RAID0 it doesn't really matter, in that case it may be better to have identical drives - if one is going to fail, everything is gone anyway. This is important for larger arrays including multiple disks - if you have 16x 2TB in RAID5, rebuilding will take long time (probably days), and then it's catastrophe if another disk fails. With RAID1 this is not major problem.


Speed: no, normal RAID controller/software RAID can't use full performance from disks with different speeds. In RAID1/RAID0 everything is striped 50% on both disks (in RAID1 mirrored equally). 50% of reads and writes goes to each disk, not depending on disk speeds.

Olli

Posted 2011-03-15T15:00:16.657

Reputation: 6 704

2RAID was originally meant "inexpensive" drives, but it has long ago ceased to be about a cheap way to use multiple drives and become a hard core way to have bullet proof redundancy. – Zeke Hansell – 2011-03-15T17:12:31.070

5Had a friend who bought an early raid system. All 5 drives seized up at exactly the same time because of a factory issue with the lubricant they used in the factory and they were all 5 from the same production batch. Don't say it can't happen. – Zeke Hansell – 2011-03-15T17:13:53.470

1@Zeke: no, I didn't say it can't happen – Olli – 2011-03-16T06:15:53.917

1+1 for talking about the speed impact that you will get from different speed drives (how fast you can read/write). – Trevor Boyd Smith – 2011-09-23T16:45:25.057