Articles: Storage
 

Bookmark and Share

(3) 
Pages: [ 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 ]

Multithreaded Read & Write Patterns

The multithreaded tests simulate a situation when there are one to four clients accessing the hard disk at the same time – the clients’ address zones do not overlap. We will discuss diagrams for a request queue of 1 as the most illustrative ones. When the queue is 2 or more requests long, the speed doesn’t depend much on the number of applications. You can also click the following links for the full results:

We’ve got interesting results at multithreaded reading. SSDs do not have read/write heads, so it shouldn’t really matter for them how many data threads they have to read. However, there is a difference as you can see. These SSDs are not so good at reading when they have to read multiple data threads. Nearly each of them suffers a 25% performance hit. The Vertex 2 comes out the winner in this test because its controller is more efficient at two and three data threads than the opponents.

We must also note one interesting fact. Every Indilinx-based SSD speeds up at four data threads. This controller must be able to effectively parallel such load to its memory access channels. This does not refer to the Colossus Lite. Its array of controllers limits the multithreaded read speed of this SSD to about 150 MBps.

Multithreaded writing is more difficult for SSDs (as opposed to HDDs that find it easier to do multithreaded writing than multithreaded reading). Almost all of our SSDs suffer a twofold performance hit when we add a second data thread for them to write, the Colossus Lite even slowing down fourfold! Interestingly, the Vertex 2 feels better in this test than its opponents even though it does not have 64 megabytes of cache that might smooth out the performance hit. It does not mind writing two data threads but has problems writing three threads. The fourth thread makes it even a little bit faster, though.

So, you shouldn’t write multiple data threads simultaneously to an SSD. It’s better to use the effective deferred writing mechanisms of modern OSes. But if you can’t avoid this kind of load, you should prefer the Vertex 2.

 
Pages: [ 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 ]

Discussion

Comments currently: 3
Discussion started: 07/02/10 04:33:00 AM
Latest comment: 07/07/10 08:17:07 PM

View comments

Add your Comment