RAID vs Separate HDD's

Question

What would be more efficient?

Let say we have 2 hdd's.

Consider this scenario:

There are 2 simulations requests.

In the RAID-0 setup the speed of the response will be boosted by the array.

In the regular 2 hdd setup. The response speed could be heavily boosted if the requested files are not on the same hdd. This way both hdds will work at their maximum capacity to serve the files.

My question is which will be faster and more efficient in the long run?

Answer 1

My question is which will be faster and more efficient in the long run?

Assuming you really want these approaches (no redundancy), you will find that unless you primarily do random access of data quantities significantly smaller than your stripe size, the RAID0 approach will perform significantly better than JBOD.

Consider the following cases:

1. Two files larger than stripe size: Chances they are on the same JBOD disk: 50%. Chances they are on the 'same' RAID disk for streaming transfer purposes: 0%

2. One file larger than the stripe size: Chances it is on the same JBOD disk: 100%. Chances it is on the 'same' RAID disk for streaming transfer purposes: 0%

3. Three files larger than the stripe size: Chances they are evenly distributed between the disks with JBOD: 0%. Chances they are evenly distributed between the RAID disks: 100%.

4. Large # of files larger than the stripe size: Functionally equivalent between the options.

5. Large # of small files / random small requests: Functionally equivalent between the options with the caveat that if the seeks are to a single file, RAID0 wins again.

The situation where JBOD wins is if you want to be able to recover some data if a disk fails. If you are using either of these methods and want to be able to recover data in case of a drive failure, you are probably doing it wrong (IMHO).

Answer 2

Know your I/O. It depends on your expected I/O patterns, as well as your failure tolerance. Which will be faster and more efficient in the long run? It entirely depends. There is no hard and fast standard.

If you have two very different workloads on that server then having each workload on their own spindle will give you the best gains... maybe. It could be that the total I/O of both combined would not exceed what could be gained with RAID0. Or maybe it will, especially if one of them involves significant number of sequential accesses where any kind of random I/O will slow things down.

The down side of RAID0 is that the failure of a single drive means all the data are gone not just half of it (technically yes, half it of it is still there, but it isn't useful). You must balance this potential with your performance goals.

Answer 3

Considering that, with RAID 0, one drive tanking will destroy all of your data ... The most efficient solution, in this scenario, is JBOD.

RAID 0 is almost never the right answer.

Answer 4

It should depend on the size or continuous nature of the read in question? RAID0 will speed up continuous read throughput but random access usually even takes a slight hit.