5400 RPM vs. 7200 RPM 2.5" Drives in RAID - Any Real Difference?

Question

Suppose you setup 8 drives in a RAID 10 configuration using 2.5" drives using 500GB capacity each (total 2TB with RAID10 config 1TB useable).

The controller (assume a typical 3ware SATA controller) is housed within a NAS setup that will use iSCSI and LVM cluster for data management/delivery over a 2GB dual nic bonded uplink.

The specs for latency on both 7200 and 5400 RPM drives are very similar. Average Latency: 5.5 ms

In terms of type of data access, consider it for server use for virtual servers hosting.

The output/throughput is being run through the RAID to give a combined total of X number of MB/s that can be delivered, so I tend to think that the RPM speed will not provide a big performance hit with the combined output of the disks.

If you are running with a RAID 10 array like this, would you probably get similar performance from both a 7200 RPM set of drives and a 5400 RPM set of drives?

Solution 1

Spindle speed does not matter, for writes, if your RAID controller "assume a typical 3ware SATA controller" is assumed to be one with a fair amount of RAM.

High end 3Ware controllers actually have a setting to flush the write back cache before returning

http://www.3ware.com/support/UserDocs/7.5.3_UG.pdf

in addition to HBA ram, your hard disk also has a write cache.

Spindle speed is more of a factor for reads and with a RAID-1 or RAID-10, the performance will benefit from load balanced reads. Thus, I doubt you'll see much difference between two 5400 vs. two 7200s in this situation.

Solution 2

While the head-movement latency is very similar, you will see differing performance from 7200 and 5400 RPM drives. The faster spin means that there will be a very small difference in the length of time that the heads wait for the right part of the track to be underneath once they are in the right position away from the spindle, and once they start reading then the blocks will pass by faster and so be read faster with the faster spinning drive giving a noticeable boost to peak transfer speeds.

Of course that last point assumes that the information density on each track is the same between the drives, which is not necessarily the case as two drives with the same total data capacity could have different numbers of platters, so you'll need to compare the drive options in a little more detail to get an answer you are more confident is correct.

Solution 3

In the setup you described you're going to end up with a pretty random read/write IO workload. So I think you'll see a benefit from the higher IOPS that the faster drives are going to get you.

If we limit this discussion to just the drives there are two main factors that influence the performance in these situations. Drive rotational speed and the number of drives involved in any IO operation. There are other factors but these two play the biggest role. This is also one of the reasons why enterprise class drives are typically smaller and run at 10k or 15k rpm. This plays into addressing both the rotational speed and number of spindles involved in any IO.

Solution 4

If your primary concern is streaming (primarily sequential IO) throughput then the 7200rpm drives will almost certainly deliver higher sustained throughput by a significant margin but a lot depends on the actual drives you choose. Random IO will probably not differ too much if, as you say, the average latency really is similar on both drive types but you really should confirm that by looking for some independent tests of the drives.

That said the most likely limiting factor for streaming throughput will be your network interface. A half decent 7200rpm hard drive can push out 40-80Megabyte/sec for sequential reads, so even though your RAID 10 setup could get 160-320Megabyte/sec sequential reads from the drives your 2GB network link will top out at the bottom end of that even under ideal circumstances and in the real world you'd be doing well to break a sustained 100Megabyte/sec.

View more solutions

Comments

[JonathanLIVE]

I understand it spins faster than therefore accesses data faster. However, the reason for my question is that I think once you have a RAID array and you have several disks now working in the data read/writes then the 33% difference in a single disk spin. Your answer doesn't seem to take the RAID and multiple disks and its function into account currently. If this was just a disk to disk comparison I would agree.

[JonathanLIVE]

Again I understand a disk to disk comparison gives no competition... but the nature of my question was to not limit the factors to just the disks.. but to take the RAID array, number of disks, and the NAS connection all together. All things considered, will I see a performance hit with 5400 RPM drives compared to 7200?

[Helvick]

The RAID setup will provide you with some multiple of the basic performance of the individual drive but the differences between the two will still be evident - if the basic 3.5" drives have a faster read (or write) rate then the RAID pack of them will also be faster by the same relative amount. Unless the controller isn't up to the task of scaling properly at any rate.

[3dinfluence]

You're question is asking for raw MB/s throughput. And if that's all you care about I don't think you'll see much difference between the two assuming that your RAID array is serving a single sequential IO stream. But being the SAN/NAS backend for multiple virtual machines that scenarios will probably never be realized. Continued in next comment.

[3dinfluence]

I think you should be concerned about how many IOPS it deliver in a random read/write workload. For this faster disks will make a difference. Like I said there are many factors and cache on the raid controller and ram in the machine can help to a degree if the working set is small enough or for short bursts of IO. For long IO operations the data still has to get to the disks at some point and the cache will eventually fill up leaving you with the raw performance of the underlying drives. At which point the cache will still help optimize the scheduling of the read/writes on the array.