SSD RAID 5 or 10?

I have a few extra SSD drives (I know its rare but this has actually happened). I figured since I have almost 500 GB in SSD's I might as well replace all of my standard HDD's with them. I currently have a RAID 5 setup and enjoy the the speed and redundancy with 10k RPM drives. Should I bother to put my SSD's in RAID 5? I have not heard of anyone doing this and have no idea how it is going to work out. The loss of space doesn't really bother me since I have a 2 TB NAS with all my media on it. Does anyone have experience with putting solid state drives in a RAID 5. If RAID 5 is a bad idea has anyone had experience with a RAID 10? is it possible to stripe the SSD's and provide redundancy with standard HDD's?

Supercereal

Posted 2011-01-12T14:01:13.157

Reputation: 8 643

I should also note that this is a subjective question (it's asking for an opinion), and these are frowned upon by the FAQ: http://superuser.com/faq -- No upvote for you, sorry.

– qJake – 2011-01-12T15:02:37.527

6@spikex I don't really want your upvote. My objective is to replace my current hdd array with sdd's I don't understand how that is not an objective question. I hope this gave you the self righteous feeling you were looking for because your post is utterly irrelevant. – Supercereal – 2011-01-12T15:23:00.933

Oh and you should probably read this http://blog.stackoverflow.com/2010/09/good-subjective-bad-subjective/ I follow that before asking any questions on any stack exchange site. My question fits most of those guidelines.

– Supercereal – 2011-01-12T15:33:59.810

I think that aside from a 100-rep user scolding and teaching a 7K-rep user how SU works, the bigger insult is that SpikeX chose to bother to take the time to answer the question before accusing it of being off-topic and not worthwhile.

– Synetech – 2012-11-01T13:36:05.857

Answers

I'd be careful with RAID 5, as any minor write requires reading, recalculating, and writing the entire block. For an SSD, this is OK if the Parity block size is the same as the SSD block size (or a multiple thereof), and they are aligned identically, of course.

If not, every write will touch two or more SSD blocks on every disk, putting more pressure on the one known weakness of SSD's - number of writes.

@Spike (can't comment yet, so it goes here)

"1000% faster" is a tad misleading. Random access time is even a bit better compared to typical drives, but sustained data rate is far from that factor. The fresh out-of-the-box OCZ Vertex 2 on my desk does 231MB/s burst and 172MB/s average read in HDTach. There's still enough headroom to the SATA II limit (375MB bus bandwidth) to allow a 2 Disk RAID.

peterchen

Posted 2011-01-12T14:01:13.157

Reputation: 1 602

1st, SATA uses an 8b/10b encoding on the wire, so actual theoretical maximum data throughput is 300 MB/s without taking any protocol overhead into account. 260 MB/s seems to be the interface limit, as Anand's Vertex 3 Pro Review shows. 2nd, 3 Gb/s is per connector, though other interfaces (DMI, PCIe) might limit overall throughput. 3rd, "1000% faster" is only an order of magnitude, which is accurate for an SSD's random I/O. Finally, find me a HDD that has >150MB/s sequential transfer rates. :-)

– afrazier – 2011-02-22T21:36:53.200

@afraizer: Yes, I underestimated the bus overhead, I've adjusted the reply. BTW this beast seems to be close to 150MB/s: http://www.tweaktown.com/reviews/3223/western_digital_velociraptor_600gb_sata_6g_hard_drive/index4.html - My experience with the 300GB model are good so far, and it's still more space for the buck.

– peterchen – 2011-02-23T08:56:01.103

Intel SSD FAQ

Q: Which RAID levels are optimal with SSD's?

A: In most cases RAID-5 will be optimal choice for server applications, providing well balanced combination of performance, capacity, data availability, and cost. More details on other RAID levels are provided below:

RAID-10: Can be used in business-critical applications that require the highest data availability and the highest write performance. RAID-10 can tolerate more than a single drive failure. However, benefits of using RAID-10 with SSD's are substantially smaller than with HDD's.

RAID-6: Not likely to be used with SSD's. Designed primarily for use with high-capacity HDDs that have long rebuild times.

RAID-1: Can be used when capacity of a single SSD is sufficient and when higher data availability and/or double the read performance are needed.

RAID-0: Can be used with workstation type of applications when potential data loss is not critical. Provides the highest performance and the highest capacity per dollar.

HAL9000

Posted 2011-01-12T14:01:13.157

Reputation: 161

In contrast to what SpikeX writes, Raid5 DOES provide redundancy. 1 drive out of 3 can fail and the array still works.

RAID10 can be notable faster for reading and MUCH fast for writing. I think the decision depends on what your needs are and if you use a hardware RAID controller. If you need fast write access, prefer RAID10. If you use software RAID prefer RAID10, too, because your system will have a high load just by calculating the difference for the level 5.

Daniel Böhmer

Posted 2011-01-12T14:01:13.157

Reputation: 585

2The only downside of RAID-10/01 is that you lose half your capacity, so it is costly. Another thing to note is that the TRIM function does not work with RAID. – paradroid – 2011-01-12T16:20:08.303

I would prefer to use the RAID 5, I like the ability to continue in degraded mode if a drive fails. However writing is what I'm mainly concerned about. – Supercereal – 2011-01-12T17:16:02.850

Kyle: Why do you think one cannot continue work when the RAID is degraded? With RAID10 up to 2 drives can fail and you will only notice a performance decrease... (any get an alert hopefully) – Daniel Böhmer – 2011-01-14T14:37:51.887

With a RAID-10, 2 drives can fail if they're on separate stripes. If both drives from the same stripe fail, the entire array fails. – afrazier – 2011-02-22T21:26:29.413

Note that RAID10 can only survive 4 of the 6 possible two-drives-failed conditions. One one drive fails make sure you replace it ASAP in case another fails soon and it is the wrong one of the remaining three. (unless you are using 3-drive RAID10 (called RAID1E with IBM controllers) in which case only one drive can fail as with RAID5) – David Spillett – 2011-02-22T21:31:36.087

The problem with raid 1,3,4,5 etc. on SSDs, is that all SSDs are aged at the same rate, and the chance of all drives failing at once is highly probable. Look up "Differential RAID", that might interest you.

Differential RAID (PDF)

Brent

Posted 2011-01-12T14:01:13.157

Reputation: 1

1Got a citation for that? – Journeyman Geek – 2012-11-01T03:39:47.010

-5

SSDs are much faster than traditional platter-based disks, so having them in a RAID, depending on how you set it up, might not be as beneficial (you'd hit the limit for data transfer rates since SSDs can be up to 1000% faster than standard disks).

If you want speed and redundancy, I would go with RAID0+1 (or 1+0), or RAID5+1 if you have enough drives. RAID5 probably wouldn't be useful since there's no redundancy.

qJake

Posted 2011-01-12T14:01:13.157

Reputation: 261

1http://www.pcguide.com/ref/hdd/perf/raid/levels/singleLevel5-c.html That should help you understand my reasoning behind this. – Supercereal – 2011-01-12T15:37:31.070

RAID5 has no redundancy? RAID0 is the only level without redundancy, as you have learnt by now ;) – paradroid – 2011-05-19T06:33:57.597