Raid SSD or Single Large Drive?

[DrakeConnar]

I was wondering which is better?

Is there a performance increase to running a raid on two smaller SSDs, or should I just get one larger SSD if prices are comparable?

Example,

Raid 0 2 64gb SSD
or Single 128GB SSD?

Also, if Raid 0 SSD is better, will two SATA II SSDs run about the same as one SATA III SSD?

Thank you.

 

[DrakeConnar]

More specifically,

I am looking at the SATA II Adata raid 0 from the $2000 Tom's Hardware Performance build vs the Vertex 3. Is the extra cost worth it?

Thank you.

 

[tecmo34]

I would go with the Vertex 3. I'm not a huge fan of RAID0, due to the potential ease of data loss. If one drive has an issue, you loss everything. Also, the speed of the SATA II RAID0 won't be much better (if at all) than the single OCZ Vertex 3 on a SATA III port.

 

[sminlal]

The flash memory chips in an SSD are organized internally as parallel channels much like RAID. When a manufacturer offers SSDs of various sizes using the same controller, the larger sizes usually have more chips ganged up in parallel, resulting in better transfer rates. Because of this you're usually better off just buying a larger SSD rather than having to deal with the hassles of RAID.

 

[geofelt]

Short answer: a single larger ssd is better.

Longer explanation.
1) siminlal and mayankleoboy have good points ^
2) The value of raid-0 is for large sequential data transfers. That is not something we do much of, except in synthetic benchmarks.
What the OS does is lots of small random reads and writes. That is a task that a ssd does very well. You will get no raid-0 benefit on that activity because the i/o will be done only to the single ssd that has the data.
3) With data on two raid-0 drives, you will lose all of your data if either drive fails. That is not a very frequent occurrence, and probably not worth worrying about. You DO have external backup, don't you?
4) If you have a sandy bridge cpu with 6gb sata, then perhaps the gen3 drives like the Intel 510 or vertex3 might be worth it. They can be limited by a sata 2 connection on sequential operations.
If you are on a budget, the extra cost of the newer gen 3 ssd's may not be worth it, again, because of how little sequential work we do.

 

[groberts101]

see.. I'm on the other side here. I've run R0 on many systems without issue for years and would never consider using just one drive. In fact this system alone runs 4 R0's.

And while Geofelt has a valid point above, he would be incorrect in some of the assumptions made(and yes I remember the conversations quite a while back about you liking a single drive, lol).

First off.. yes.. there are better nand/IC configs in larger drives. BUT.. the simple fact is that 2 smaller drives in R0(2 controllers running 16 channels) will always trump a larger single drive(1 controller running 8 channels) regardless of the nand chips density config.

Everything just increases all around with more channels. Also consider that Trim does not recover speed on Sandforce based drives due to Durawrite, and that's a non-arguement as well. Then there's the increase in performance when writing incompressible data(when discussing Sandfocre controlled drives). A single larger drive will not hold a candle to 2 smaller ones in raid and goes far beyond what any benchmark will tell you in relation to OS duty.

Typically, the ones that don't see the benefits from raids are the ones who never really needed it with their usage levels in the first place. I've converted many to SSD raids and if you task your system hard enough to use it?.. you'll not go back to single drives either.

Having a Vertex 3 to beta test has also shown that I would still prefer 2 slower Vertex 2/120GB over one faster Vertex 3/240GB for heavy usage. While sequential speeds using easily compressed data is similar.. the results change as you move towards more incompressible data streams and heavier tasking. SSD is very much about channel count and raids make it quite apparent to those that can utilize it.

 

[JohnnyLucky]

Personally I prefer a single large capacity ssd over a Raid array.

 

[geofelt]

Some time back, I had a intel x25-M gen1 80gb ssd. It worked well. I needed some more space for the OS drive, and did not want to manage two drives. So preferring a single image, I got a second X25-M 80gb drive and converted to a raid-0 with a 160gb single image. It ran fine, and sequential benchmarks looked good. In truth, though, I saw no improvement in everyday opertions. My son needed a ssd, so I gave him one, and sold the other, replacing the raid-0 with a single X25-M 160gb gen2 drive. My perception was that it was a smidgeon faster. Probably because of the gen2.

I think the problem with SSD benchmarks is that they push the drives to limits that the normal user never sees. They measure throughput, often at very high queue levels.

What would be more useful would be a measurement of response times at low queue levels. There are some statistics out there, and an unscientific reading says that most ssd's have about the same response times.

 

[JohnnyLucky]

^5 +1 what geofelt said.

Consider this:

1ms = 1/1,000th of a second.

A blink of an eye takes between 300ms and 400ms.

Can the human eye and brain notice the difference between 0.01ms and 100ms? Absolutely not!

 

[groberts101]

LOL. then you guys need to just try a fast SSD raid setup(not old gen 1 stuff either) one time in your life... cause seeing IS believing. You certainly don't need a stopwatch to measure it either.

Also consider the fact that Sandforce drives have ultra low latency when combined in raids unlike other controllers that will have cumulative affects for each drive added to the point that latency increases become noticable.

I don't benchmark my drives anymore as my config and tweaks are about as good as it gets for this system and would just be useless numbers on the screen along with unecessary writes to the drive. I use Windows, Adobe, and as many applications running all at once as I can possibly keep track of, to tell me that things are much faster. Again.. no eye blinks or stopwatch required.

 

[sminlal]

LOL. then you guys need to just try a fast SSD raid setup(not old gen 1 stuff either) one time in your life.

My system boots from hard disk from "Starting Windows" to logon screen in about 23 seconds, and in only about 10 seconds with a single, non-RAID 160GB Intel X25-M G2. The Youtube videos I've been able to find using equipment similar to mine but with two 80GB Intel X25-M G2 drives in RAID-0 boot from "Starting Windows" to the logon screen in about ... 10 seconds.

I remain unimpressed with RAID-0 for SSDs.

 

[groberts101]

I remain unimpressed with RAID-0 for SSDs.

Then it would be pretty obvious that you didn't need raid in the first place.

And measuring SSD performance with boot speeds is almost comical. I too can get my boot speeds all the way down to nothing if I run a lean install with not much starting up or get rid of all my hardware. I and most that use the computer for more than surfing, games, or MS Word usually measure performance by how much can be done in a shorter timeframe once you are actually using it. All my drives and raidcard takes way too long to initialize so I just run 24/7 which also gives the nice benefit of heavy GC time for all my SSD's.

This system is so fast compared to a single sata SSD(any SSD) that my typing speed and ability to multitask that fast are the biggest bottleneck now. Heavy multitasking ability saves time(encoding video while importing photos and streaming music with many other apps and files being read/written at once can tax a single drive pretty hard as all that work gets divy'd up from the available bandwidth) and raid increases this ability due to channel count. With ANY SSD more channels is always better. Some just don't need it and that's fine,.. but you guys shouldn't downplay things for which you haven't tested(benchmarks aside) or even need for your usage level.

I do know without a doubt that most who live with raids for a while?.. rarely go back to single drives(HDD or SSD) as the bar gets permanently raised form that point onwards. The one's who go back do so either because they didn't need the extra speed and headroom created by it, had data integrity issues, or didn't want the complexity of raid. Personally.. I don't even like booting to the V3 as my 6 drive array is so much better at just about everything. Being a long time hot rodder makes me sort of power crazed though.. so there's that. lol

 

[sminlal]

Until I can find a Youtube video that shows how SSD RAID performs while I'm doing the stuff that I usually do, boot time will have to remain a reasonable proxy.

All your nice descriptions about how fast your system is and how you can't keep up with it apply to my case as well, so why would I bother with RAID?

I've been a datacenter administrator for many, many years and RAID organizations of one flavor or another are among my standard tools. I'm not afraid of RAID, and I have plenty of experience with how it performs. But on my home system there's no practicable benefit to me in using for an OS SSD, and I'm fairly confident that's true of most typical users.

The degree of performance improvement you get going from a standard HDD to a single SSD is huge compared to what you get going from a single SSD to two in RAID-0. For most people that's more than enough, IMHO.

 

[JohnnyLucky]

^5 +1 what sminlal said.

Remember, we're talking typical home users.

As near as I can tell, solid state drives in a Raid array are overkill. From my perspective it is the equivalent of purchasing a 1,200 watt power supply for a pc system that only requires 400 watts for optimal performance. Anything over 500 watts is not necessary since it will not make a difference - unless someone wants bragging rights.

 

[sminlal]

For me it falls into the category of performance improvements you can measure, but which you won't notice in daily life. It's a truism I picked up back in the days of the original IBM PC with it's Intel 8088 processor. I eagerly purchased a pin-compatible NEC V20 CPU and installed it, expecting great things. What I discovered was that the processor was indeed faster - my timings showed about a 10% reduction in wall-clock time to do typical tasks. But the "feel" of the machine wasn't any different, and in fact the real-world improvement I was expecting really didn't materialize.

From that day on I realized that small differences in benchmark performance are largely meaningless. SSDs vs HDDs is a huge difference that really makes your PC feel a lot faster at certain things. But single SSDs vs. SSDs in RAID 0 is one of those things you can measure in a benchmark and that's about it.