What you will learn from this tip: Which RAID configuration offers the best price-to-performance ratio.
A SearchStorage.com reader, Mike Bennett, recently submitted a couple of questions about a SAN quiz that we posted back in 2003. I posed his questions to our SAN expert, Christopher Poelker, to address Bennett's concerns about the way we explained the relationship between RAID price and performance.
We try to provide the highest quality information to our readers, and reader feedback is important to us. Please feel free to contact us anytime with questions or concerns via e-mail at editor@searchstorage.com.
Read the Q&A below.
Q: I had a question about the SAN Quiz #1 dated August 13, 2003. In that quiz, question 7 asked:
The best price-performance ratio that you can get from a RAID subsystem is available from:
a. RAID-0
b. RAID-1
c. RAID-5
d. RAID-0+1 (or RAID-10)
The answer given is: RAID-0+1 (or RAID-10).
The explanation then goes on to describe each RAID type, starting with RAID-0, which is described as "...offers the best performance but no fault-tolerance". Further on, it describes RAID-0+1 as "...offers higher performance than RAID-1 but at much higher cost".
It seems to me that, since the question is asking specifically about the best price-performance, the descriptions suggest that RAID-0 offers the best performance, and since all disks are used to store data (no extra disks for parity or mirror copies), it would necessarily also have the lowest cost. This makes RAID-0 the best price-performance ratio. In addition, the comment about RAID-0+1 costing much more than RAID-1 confuses me -- in what way are costs "much higher" with RAID-0+1 than with RAID-1, if they use the same number of disks?
A: Mike is correct here... to a point, and the issue is in how the question is posed.
There are two basic design goals in using RAID technology, one is performance, and the other is data protection. RAID-0 offers the best performance since the overhead with writing parity is eliminated, and data is striped across many spindles, which causes an increase in IO's per second. Since no parity is generated, the capacity of all the disks in the RAID set is fully utilized, which generates cost savings in the number of disks required for a given capacity. The problem with RAID-0, is there is no data protection. Lose one spindle, and ALL the data in the RAID set is GONE. (Unless you back it up of course!) So, there is a trade-off with using RAID-0. Great performance, but no protection. Questions based on RAID generally assume you want to also protect your data.
RAID-1 is a mirror, usually created with two disks. With RAID-1, you lose the capacity of the mirror partner for the copy of the data. RAID-1 is excellent in data protection, but does not offer the advantages of striping across many disks. With RAID-10, you get the data protection inherent in mirroring (RAID-1), and also get the advantage of RAID-0 striping performance. It costs more, because you need more disks to create a RAID-10 stripe set (usually 4 or more). If there are 4 x 20 gigabyte (GB) disks in a RAID-10 set, that would be 80 GB of raw capacity, but the actual usable capacity is only 40 GB, since half the disks are used for the mirror copies.
So, the indicated correct answer is still valid assuming you want good performance but also want data protection. The best PRICE performance would, in my opinion, be RAID-5, since it provides data striping, and you only lose one spindle's capacity for parity. Therefore, fewer disks are required.
It is generally accepted in the data storage industry that, for database best practices, RAID-5 offers the best trade-off for price and performance with data protection. RAID-0, although offering the BEST performance, would never be used in a database environment unless the data was transitory and could be lost. RAID-10 has the best performance with data protection, and RAID-1 is used for good performance and data protection with low capacity requirements, such as database logs.
Q: If this is not correct, could you show me using sample numbers how RAID-0+1 offers better performance or price than RAID-0? Otherwise, an adjustment to the answer (or rephrasing of the question) would be appropriate so as to not confuse your readers. I understand that RAID-10 would offer the best protection-performance ratio, but since it requires double the number of drives, in no way would it offer a good price ratio compared to any of the other RAID configurations.
A: Again, RAID-0 offers the best performance and is cheap because no parity is used, but data is left at risk, so it would rarely be used for critical application data. (unless your a gambling man and like risk!)
RAID-10 is more expensive because it does use data protection, and requires twice as many disks as RAID-0 for the same usable capacity. Again, here RAID-5 would offer the best price-performance mix. I guess the answer should be changed to RAID-5, and there would be no confusion.
For more information:
RAID-50: RAID-5 with suspenders
