Better Performance: Two 128GB SSDs in RAID-0 or One 256GB SSD?

[belonky]

So, I have a Mac Mini Server with two drive bays and I want to install an SSD setup. It's my understanding that with SandForce-based drives, lower capacity drives don't perform as well as higher capacity ones, though I'm still unclear on exactly why that is (can anyone point to a good explanation of this?). I'm on somewhat of a budget constraint, and I can only afford about $500. So my questions are twofold:

1) Does RAIDing two 128GB SSDs address the specific issue causing the performance delta between 128GB and 256GB drives? I'm assuming not, since each SSD's controller doesn't see the additional addressable NAND of the other SSD. Rather, the RAID controller handles the striping.

2) That being said, I'm assuming the additional performance offered by RAID-0 more than compensates for the lowered performance of 128GB drives, and that the overall performance of two 128GB SSDs in RAID-0 would blow one 256GB SSD out of the water. Would that assumption be correct?

2a) Just out of curiosity, what is the performance delta, if any, between two 128GB SSDs in RAID-0 and two 256GB SSDs in RAID-0?

Thanks!

- Jon

 

[Hulk]

I would bet sequential read/write would be 50% or so improved but real world results, outside of giant file copies, would be not noticeable do to increase overhead of the Raid setup.

But what might be noticeable is the possible failure of a drive. SSDs are still not as reliable conventional drives and putting them into Raid 0 for anything but perhaps storing games is kind of scary in my opinion.

 

[belonky]

But of the two, two SandForce-based 128GB drives in RAID-0 or one SandForce-based 256GB drive, which would provide better performance? (I don't care about reliability; I'm diligent with backups). Thanks.

 

[Hulk]

Honestly I don't know but others here might. As I wrote I think performance would be about the same with sequential read/writes having the edge with the Raid set up.

 

[Sunburn74]

I would bet sequential read/write would be 50% or so improved but real world results, outside of giant file copies, would be not noticeable do to increase overhead of the Raid setup.

But what might be noticeable is the possible failure of a drive. SSDs are still not as reliable conventional drives and putting them into Raid 0 for anything but perhaps storing games is kind of scary in my opinion.

I've had my x-25ms in raid 0 for 2 years and they've been rock solid. In fact, I've gone through 3x seagate barracudas in that time.

Its a tough question to answer. Things seem to indicate you get performance boosts (look up vertex raid-o results). However earlier drives were less trim dependent so I can't be sure how the new ones will do. Iwish someone would just test.

 

[Hulk]

I've had my x-25ms in raid 0 for 2 years and they've been rock solid. In fact, I've gone through 3x seagate barracudas in that time.

Its a tough question to answer. Things seem to indicate you get performance boosts (look up vertex raid-o results). However earlier drives were less trim dependent so I can't be sure how the new ones will do. Iwish someone would just test.

That's good news about your drives. As you can see from my sig I have an Intel 320 so I'm hoping reliability is as good as mechanical drives. I don't even want to write this, but I haven't had a drive failure in my 25+ years of computer use. My only point was outside of anecdotal evidence, statistics would indicate SSDs are still not as reliable as conventional drives, but they do seem to be getting better with each generation. In addition it appears as though we're at a widespread tipping point for adoption now. So we should see prices really start to decline and reliability increase as the production numbers ramp up.

 

[beginner99]

Go with a single drive. Less hassle, more reliable and I guess also cheaper. But I would not touch sandforce with a 10-foot pole and even less if it is from OCZ. Get the m4 or intel 510 instead.
And who ever said platter drives are more reliable, numbers please? I very much doubt that.

 

[belonky]

Btw, just a note on reliability of SSDs in a RAID-0 config: It's my understanding (someone please correct me if you know this to be wrong) that, contrary to platter HDDs, a RAID-0 array of SSDs actually has at least the same MTBF as either one of the individual SSDs and potentially up to 2x the MTBF of either of the individual drives. This is because SSD MTBF is a direct function of the number of write cycles executed, and with RAID-0, each drive sees roughly half the number of write cycles as it would on its own. Conversely, mechanical HDDs wear at a more constant rate, due to head movement for seeks, etc. Now, of course an SSD could fail due to a non-wear-related issue, the odds of which still double in a RAID-0 configuration, but the number of failure points for SSDs is dramatically lower than there are for mechanical HHDs, since SSDs contain no moving parts. And the increased reliability from halving the number of write cycles per SSD in RAID-0 would likely offset, in aggregate odds, the increased odds of an "other" failure resulting from putting two SSDs in a RAID-0 array, resulting in a net aggregate increase in reliability.

In other words: You can't think in terms of traditional mechanical HDD paradigms when you're dealing with SSDs. But I would definitely be interested in seeing some testing to confirm this effect. Anand, are you listening?

 

[beginner99]

Btw, just a note on reliability of SSDs in a RAID-0 config: It's my understanding (someone please correct me if you know this to be wrong) that, contrary to platter HDDs, a RAID-0 array of SSDs actually has at least the same MTBF as either one of the individual SSDs and potentially up to 2x the MTBF of either of the individual drives. This is because SSD MTBF is a direct function of the number of write cycles executed, and with RAID-0, each drive sees roughly half the number of write cycles as it would on its own. Conversely, mechanical HDDs wear at a more constant rate, due to head movement for seeks, etc. Now, of course an SSD could fail due to a non-wear-related issue, the odds of which still double in a RAID-0 configuration, but the number of failure points for SSDs is dramatically lower than there are for mechanical HHDs, since SSDs contain no moving parts. And the increased reliability from halving the number of write cycles per SSD in RAID-0 would likely offset, in aggregate odds, the increased odds of an "other" failure resulting from putting two SSDs in a RAID-0 array, resulting in a net aggregate increase in reliability.

In other words: You can't think in terms of traditional mechanical HDD paradigms when you're dealing with SSDs. But I would definitely be interested in seeing some testing to confirm this effect. Anand, are you listening?

Writes are excatly the same (ignoring possible raid-0 overhead) for 2 128 GB and 1 256 gb drive because both setup have the exact same amount of flash cells.

MTBF is MTBF and if you combine 2 devices with MTBF of 1 they combined MTBF will be 0.5 regardless of device type.

While you gain speed you lose reliability and TRIM. Personally raiding ssd in a home setup is useless except for the fun of it and e-pen-factor.

 

[nanaki333]

I've had my x-25ms in raid 0 for 2 years and they've been rock solid. In fact, I've gone through 3x seagate barracudas in that time.

Its a tough question to answer. Things seem to indicate you get performance boosts (look up vertex raid-o results). However earlier drives were less trim dependent so I can't be sure how the new ones will do. Iwish someone would just test.

i had my intel G1 160s in raid0 for over 2 years using the same install the whole time. never had an issue. switched to 320 160s for raid0 and still have had no problems

 

[TemjinGold]

OP: The reason bigger drives are faster is that an SSD is actually a RAID of the individual nand chips themselves. Bigger drives have more chips, so it's like RAIDing more drives together. Therefore, 2x 128 vs. 1x 256 will actually be fairly similar in speed (all else equal) but the 2x solution does not have TRIM and is more hassle. The tradeoff is that the 2x solution is usually a bit cheaper. If money is not an issue for you, the 1x solution is better in pretty much every way in terms of real world usage.

 

[groberts101]

2 medium drives will be faster than 1larger drive.

I've run 4 different controllers so far and all give the same benefits with increased multitasking performance. Greater caching benefits is also an added side effect for any Intel based raid array.

Does the array start Windows any faster? probably not

Does the array let you navigate the OS's GUI any quicker or open an app? probably not

Does the array let you open a file that's natively stored on the volume any faster? probably not

Does the array let you write at higher speeds when combined with a storage volume that is able to read at least that fast? you betcha!

Does the array let you work faster overall while you encode vids, surf the web with 12 browser tabs open while also DL'ing/streaming media, and then to simultaniously copy a folder full of music or pictures all at the same time? you can bet yer single little SSD ass that it does!

raid works well with Sandforce drives and all that's needed over the longterm is enough free space reserve combined with an ocassional logged off session to allow garbage collection(GC) to optimize the drive/s internal structures.

TRIM is WAYYY over-rated with these controllers due to how they most commonly utilize those trim marked blocks during GC recovery time.

 

[Burner27]

Here is a 3 drive RAID 0 array using 3 x 128GB Crucial C300s:

 

[Shmee]

Impressive!

OP, which sandforce ssds were you considering? Also, what controller would they be on? I assume you are using Mac OS?

If it can wait, you might also want to consider the samsung 870 when it comes out.

 

[deimos3428]

I have run a 3x60GB RAID0 for nearly a year. While it benchmarks faster than a single drive, the real-world performance increase is negligible. My recommendation is to base your purchase on pricing; unless the RAID configuration is significantly cheaper go with the single drive.

 

[groberts101]

I have run a 3x60GB RAID0 for nearly a year. While it benchmarks faster than a single drive, the real-world performance increase is negligible. My recommendation is to base your purchase on pricing; unless the RAID configuration is significantly cheaper go with the single drive.

well.. not to say that you don't know what you're talking about but youre usage levels obviously reflect only the first 3 questions I mentioned above.


also the OP should consider that the write speeds of incompressible data are slower with any Sandforce controlled drive due to the loss of internal compression. So again.. 2 drives WILL be faster than 1 larger drive in that respect as well.

If it's all about price and ease of implementation then I'd agree 100% that a single larger cap drive is fine.

If it's about perofrmance though?.. raid is the way to go and the harder you push it.. the faster it will seem compared to a single drive.

I absolutley hate to boot a single SSD anymore these days. 6 Sandforce drives in R0 tends to have that effect. lol