Single SSD or two SSD's in RAID0?/Benchmark results

[antares]

Hi, I just changed my system from 2 SSD's in RAID0 (Samsung 830 Series 256MB each) to a single SSD (Samsung 840 Pro 512MB). The reason I changed it is mainly because I need to encrypt the whole volume with Truecrypt which does not support RAID.
I measured the performance in both configurations, here are the results:

So what do these results say? I can't seem to tell the difference, I tried different applications and they open almost instantly in both cases. Boot time seems a bit faster with the single SSD (I guess that's because of the software RAID initializing at startup?).

 

[Lady Fitzgerald]

...Boot time seems a bit faster with the single SSD (I guess that's because of the software RAID initializing at startup?).

Most likely. Many people don't realize that, despite impressive bench results, in real life, there is little to be gained when putting newer SSDs into RAID 0. Add to that reduced reliability; if one SSD dies, the whole array is lost and, since there is more than one SDD, the chance of failure increases with each added SSD. Still, I see people wanting to put several SSDs into RAID 0.

RAID 0 made sense with slower spinners; the delay caused by the RAID initializing was negligible compared to the overall gain. SSDs, however, are so fast, any gain in speed will be smaller since the overhead delay stays the same and will offset a larger percentage of any gain from the RAID.

Keep in mind a faster boot drive will decrease only boot time and program loading. If you have a program that takes one or two seconds to load with one SSD, cutting that down to 1/2 to one second just isn't going to be particularly noticeable.

 

[LMiller7]

In any computer system there will always be a bottleneck, something that is limiting some aspect of performance than any other. Upgrading that something will not eliminate the bottleneck but simply move it elsewhere. Typically the component limiting boot and application load times will be a slow conventional drive. Upgrading to an SSD either eliminated or greatly reduced the influence of disk performance nfor this aspect of performance. But you can't expect that further improvements in SSD speed will noticeably improve performance. It has ceased being the bottleneck.

RAID 0 can produce some impressive benchmarks, one of the most noticeable being the sequential read times. That is the time for reading large blocks from large files. But that doesn't happen very often in the real world where small reads are more typical. Here the performance of RAID 0 is more modest.

System performance is a complex thing with multiple factors involved. Upgrading one component often doesn't provide the performance that the benchmarks might imply.

 

[Britton30]

Here's a single 840EVO with Rapid Mode enabled.

 

[antares]

Amazing benchmarks Britton! Congrats!!. I downloaded Samsung Magician software and set it to "Performance" before posting my benchmark. The Rapid Mode is not yet supported for the Pro version, only for the EVO, however it seems it's coming later this year for the Pro too.

 

[Britton30]

Thanks Antares, I got the EVO just to try out the RAPID mode. Hopefully Samsung gets busy with the Pro version, the year is nearly over.

 

[antares]

hi Britton, benchmark aside, can you feel a performance difference in Rapid Mode compared to standard mode? Is the difference subtle or big?

 

[Britton30]

No I don't notice any speed gain. It is in the testing phase though so I don't yet have resource heavy software, like Photoshop and Corel Draw installed. I think it's like overclocking, there's no performance gain but gives good benchmark numbers to brag about.

 

[Lady Fitzgerald]

From what I've read, it uses system RAM to cache the SSD, similar in effect to RAID 0.

 

[Britton30]

YEs it uses up to 1GB of RAM and plans are to make it adjustable to use more.

 

[Lady Fitzgerald]

Now that might justify getting another 32GB.

 

[Britton30]

Sure it would and one can run VMs with more allocated RAM.

 

[Lady Fitzgerald]

Big VMs.