SSD RAID scaling under Windows 7 @ [H]

[Kyle_Bennett]

SSD RAID scaling under Windows 7 (http://www.hardocp.com/article.html?art=MTY1OCwxLCxoZW50aHVzaWFzdA) - What can you expect when you get your new Windows 7 install working with those crisp new solid state hard drives? SSD drives are not what they used to be already. And just how is that Intel ICH10R chipset serving you?

One final chart before we go. What will running all these drives in RAID-0 get you in terms of boot up speed? Do we finally have an instant-on PC?

[qdemn7]

Been using a single 60GB OCZ Vertex with XP on a X38 chipset DFI board for the last few months. Even with the older technology, there's no way in hell I would go back to using a standard hard drive as my OS drive.

[kuyaglen]

I gotta get more sleep, it took me a closer look to realize the bootup chart was in seconds. But was the time started from when turning on the computer thus also taking the time for the controller's to load up the array or was the time taken when the Windows 7 loading screen came up?

I would of liked to see how game loading performance also...but you had 4 drives when these people had like 26 link (http://www.engadget.com/2009/03/09/24-samsung-ssds-get-strung-together-for-supercomputer-fun/).

One thing also about ssd's that I like is the weight saved...4x 3.5" hdd weigh a whole lot more than 4x ssd's.

[extide]

Based on my experience with my SSD it seems like device initialization now takes up a large percentage of time, as opposed to waiting on the disk most of the time.

[Methadras]

Once you go SSD you never go back. The upsides far far far outweigh the downsides and there aren't that many downsides because they can only get better because they aren't mechanically based anymore.

[vengence]

Interesting results. Very different from what I would have thought. So what I'm hearing is after I win the uber lotto (1Billion+ USD), I should run a single SSD for my OS and I really only need a 3x raid for programs.

[tricky0502]

did any of those raid setups reach 7.9 on the wei?

[BrndNtrl]

While the upsides are getting better, I think I'd still much rather wait a few years for SSDs to mature and go for larger capacities on HDDs. There are simply too many variables in buying them for me still (price, performance scaling poorly with space used, overall capacity, etc).

I'd love to have an instant on PC, but waiting an extra 15 seconds for windows to load to avoid having to delete/reinstall games every week is time I'm willing to spend.

[Dekoth]

I have plans on making the leap to SSD around early next year, so this article was really informative for me. I was still torn as to if I should go with a single internal ssd and convert my current raid to an external storage or if I should go all out will a full on internal SSD Raid. From the looks of things given how infrequent I need to access what would be on the external storage, it looks like I can save the extra and opt for a single internal.

[Pretiacruento]

_Great_ review, guys! I hear ya Dekoth, we're in the same boat ;) and this article was very, very informative (very n00b friendly even for a guy like me). That's why I LOVE this page! :D

But was the time started from when turning on the computer thus also taking the time for the controller's to load up the array or was the time taken when the Windows 7 loading screen came up?

Yep, Kyle said it measured the time from the moment W7 is starting up, to remove the BIOS overhead.

[SamuraiInBlack]

While the upsides are getting better, I think I'd still much rather wait a few years for SSDs to mature and go for larger capacities on HDDs. There are simply too many variables in buying them for me still (price, performance scaling poorly with space used, overall capacity, etc).

I'd love to have an instant on PC, but waiting an extra 15 seconds for windows to load to avoid having to delete/reinstall games every week is time I'm willing to spend.

I'm with you on that. I try not to be an early adopter of technology because I end up getting bit in the ass for it. So I'm going to wait until things get sorted out. That, and I have a lot of data to store between games, music, and movie rips, and I'd rather buy a single terabyte hard drive rather than shell out the cash for a terabyte worth of SSD's. I need the capacity more than I need raw speed.

Maybe once the technology matures more and there's decent capacities (500+) at decent prices, I'll hop onto SSD for a strictly gaming rig.

[mikeblas]

The increased overhead of RAID also makes it a slightly slower option when compared to a single drive in our case. Huh? What does that mean, specifically?

[grossebeaver]

I'd be curious to see what the same tests would yield using a pair of SLC based SSDs. Particularly with the write speeds.

[Digital Viper-X-]

Huh? What does that mean, specifically?

Basically, raid 0 splits up the data into 128kb chunks and sends it out 128kb at a time to the different drives, while this may help throughput, it can also hurt seek time. specially with SSDs which have virtually 0 seek time, now you're introducing the time it takes the controller to do the work too.

I could be wrong :P

[evilsofa]

Editing the introduction for grammar and flow. I'm sorry if the edits look messy, but I can't figure out how to do strikethrough on this forum.

Most of the time when we start a new article here at the [H]{missing comma} it goes pretty much as planned{missing comma} with the new hardware beating the old stuff by some margin. Sometimes articles take on a whole new shape as{replace "as" with "when"} your results don’t line up with expectations. Even{replace ". Even" with ", and"} the technology being tested can improve in the short time between the start of an article and its publication. Working with cutting edge technology{missing comma} it is always a little tricky knowing if{replace "if" with "whether"} the results you’re seeing are valid or {insert "if they are"} some aberration from drivers or hardware incompatibility. SSDs are the prime example. Anyone looking to purchase an SSD should know that the technology behind these drives is progressing at such a fast pace that what was a{remove "a"} practical knowledge yesterday is out of day by{replace "out of day by" with "obselete information"} tomorrow.

In case you don’t know where I am going with this{missing comma} let me be frank. When this article was conceived{missing comma} it was to be an informative article for both newcomers and enthusiasts to the world of SSDs. What{replace "What" with "We wanted to explain what"} steps were needed to properly set up your new SSD{missing comma} as well as how to get the very most{replace "most" with "best"} performance from your drive. As mentioned before, the technology behind SSD has progressed so quickly that much of the information out on the web is on the verge of being obsolete, save for some older SSDs.

For today's article{missing comma} we are going to show you guys what a lot of{replace "guys what a lot of" with "what"} you can expect when you jump on the Windows 7 bandwagon and slap together some RAID-0 SSD awesomeness with a sprinkle of the Intel ICH10R Southbridge aka {insert "the"} x58 chipset. Just for good measure{missing comma} we are going to toss in some {insert "performance"} charts when running a dedicated RAID card to see what, if any,{replace ", if any," with "kind of performance"} boost can be expected from such a setup.

[greenfrogman]

if the SSD is an 200MB/s device (and is not JMicron) under nomr use not lieky to see the improvement of SSD in raid, unless you got 4 SSDs with 2 RAID 0 setups so you can watch 4gb files transfer at 300-400MB/s or you run benchmarks get big numbers but not real world

i got an Corsair S128 considering the Read and Write is 90MB/s / 70MB/s every thing responds fast even games, now the P128 is out (seems to of replaced the S128 now as not in stock any more) but under norm use my games mite load 1-3 secs faster then my SSD

if you going with SSD get an second gen SSD and get the size you need as when windows 7 comes out there be an update that give TRIM support for windows 7 and that should keep the drive at peek performance (Trim may not work on RAID)

[aggiec05]

while I like the article, it still feels a bit meaningless when none of this is entry level or even affordable for the average consumer, maybe thats just me but 4X $300 drives and a $400 raid card to keep it comparable..... wait I forget gotta shell out to keep up with the times

[mikeblas]

I agree about the meaningless comment -- it should include spinning storage as a baseline for comparison.

[MT_Head]

"In our brief overview of RAID scaling under Windows 7 we showed you the limit that the popular ICH10R Southbridge places and yours PC's drive throughput."

Should be "places ON your PC's drive throughput." I hate to be that guy, but I had to read this sentence three times before I understood it.

[fps4ever]

while I like the article, it still feels a bit meaningless when none of this is entry level or even affordable for the average consumer, maybe thats just me but 4X $300 drives and a $400 raid card to keep it comparable..... wait I forget gotta shell out to keep up with the times

Agreed...total cost for faster boot/seeks times is ridiculous at this stage of the game. You can purchase a complete high end gaming machine with monitor OR pay for a 4x SSD hard drive raid setup w/raid card for the same price. Seems like a no brainer to me. SSD's will mature, get better and get cheaper so no harm in waiting for a while...

[player-x]

I would love to see this test again using a top ARECA controller.

For example the ARC-1680ix-08, i know its not a cheap card +/- $500, but i upgraded my server at home to a ARC-1680ix-16 from a SuperTrak EX16300

I minimal tripled my true put on big files it was even up to 5x times.

I know $500 isn't in the budget range of most users still it would be interesting to see if a top controller can really make the different and see if its worth the money on these extreme fast SSDs

[vick1000]

The only problem is severe degredation of perfromance with use and volume of NAND based SSDs...

http://techreport.com/articles.x/16979

...the Intel based units do pretty well, but they still degrade unreasonably and are by far the most expensive.

[player-x]

The only problem is severe degradation of performance with use and volume of NAND based SSDs...
The new SSDs based am the new samsung controller don't have this problem any more ore at least a lot less

[GoodOlBoy]

Would an SSD allow me to scroll through The [H]ard|OCP home page without a 10 second delay while my computor digests all the fancy ads? ;)

[Citizen86]

800 mb/s..... that's fast.......

[burthold]

There are a few other things to take into account.

Highpoint is entry level to be sure, they are using the latest IOP from Intel but only running it at 800Mhz does gimp it some.

I though your price on the 4310 was high, I found it on eWiz for like 300.00, but hey what is a hundred bucks when you have already bought 4 SSD's.

I agree Arcea is a better card all the way around even on the previous gen models, but you get what you pay for.

If you are running a RAID 5 array the on board is going to get pounded by the hardware card. Since you are doing RAID 0 there is no calculation overhead for parody.

Also block size you formatted the file system in has a large impact on performance. Most SSD's read well at one block size and write well at another. 32KB to 64KB for reads and 2KB to 4KB for writes is a good point on most that I've worked with.I may may have missed it but I didn't see what block size you had on the file system and if it was sector aligned at all.

-wes

[swatbat]

Interesting. I'd love to see what happens when you try to run a raid 5 or 10 using the drives.

Also any spikes in cpu utilization using the onboard controller?

[bigdogchris]

Am I foolish to believe that for a machine that's just used for Internet and gaming, Raid really doesn't provide any performance increases as long as you have a newer, fast, single magnetic drive?

[mikeblas]

Basically, raid 0 splits up the data into 128kb chunks and sends it out 128kb at a time to the different drives, while this may help throughput, it can also hurt seek time. specially with SSDs which have virtually 0 seek time, now you're introducing the time it takes the controller to do the work too.The overhead only appears if the request size is smaller than the stripe size, then. And the overhead you're describing would seem negligible anyway; the controller just does a little work, and sends the commands along. They execute concurrently.

I could be wrong :POr you could be right. There's lots of room for guessing when trying to make sense of such a vague statement.

[Earp]

Seriously, who's going to run 4 drives in RAID 0? Unless, of course, you like a really low MTBF. I don't know how SSD drives compare with rotating rust on reliability, but I'd never chance it on hard disks.

I'd also question calling the Highpoint 4xxx series entry level raid cards. Maybe for an enterpise system, but for home use, I'd consider the 2xxx series entry level. Especially with only RAID 0, where you don't have to calculate parity, the dedicated IOP on the 4xxx series doesn't do you a lot of good.

Unlike your average integrated RAID, most dedicated RAID cards will support online capacity expansion, allowing you to add a drive to the array and have the array still be usable with the OS running while the card reconfigures the drives (which can take a while). The can also handle hot spares, email & audible alerts if a drive fails, but I suppose that doesn't matter if you're running RAID 0, you'll find out when nothing works.

While it may not be true for Intel chipsets, I've always found the drivers for NVIDIA SATA RAID to be really flakey, if you could get them to work at all. I've never had driver issues with dedicated disc controllers (both RAID & other).

I have two systems with Highpoint 2xxx series cards (a 2320 & 2310) and couldn't be happier. I bought them solely for online capacity expansion and to avoid driver issues. I'd never use an integrated RAID controller for anything other than RAID 0 or 1 on a pair of drives.

[MacGyver467]

This article couldn't have come at a better time for me. I recently purchased a 128GB Patriot Warp SSD. I'm thouroughly amazed at the performance difference between my WD 250AAKS and this drive. I had the option of grabbing a 64GB SSD, but with the advent of "large footprint" OSs (Vista: 15GB, 7: 20GB), I thought it would be best to invest in the larger size. Like I mentioned, the performance increase is noticeable.

To load a TF2 map take seconds compared to a full minute with my WD AAKS drive. Other games show massive level load time decreases as well.

To make my post relevant, I was considering purchasing a second 128GB Warp and running them in RAID0. Yes, I see the sequential transfers increase proportionally with each additional SSD added, but the increase in OS boot time shows me the random read performance has something to be desired. Adding more SSDs would only benefit a company running a large database, and not the average user.

[extide]

It would be interesting to see those boot time benches re-done at a really small stripe size, like 4KB.

[Alpha736]

Since this was an SSD/RAID preview for Windows 7, I'm sorta wondering why there weren't benchmarks done comparing it to Vista. I'm curious as to how much faster Windows 7 would be in respect to SSDs.

BTW, I haven't been keeping track of these things as much as I used to, but has Microsoft made optimizations specifically for SSDs in Windows 7?

Otherwise, it was a great review and interesting read! :D

[AceGoober]

Thanks for the review, [H].

It's going to be a while before I purchase any SSDs. I'm holding out until the product matures more before jumping on the bandwagon. That means I'll have to continue wearing this bib to catch the drool everytime I see benchmarks performed. :p

[Digital Viper-X-]

Im running 3 Vertex drives in raid 0

Windows 7 = very very nice with that setup

[Robstar]

What about comparing to mature software raid like mdadm ?

[mikeblas]

Seriously, who's going to run 4 drives in RAID 0? Unless, of course, you like a really low MTBF. I don't know how SSD drives compare with rotating rust on reliability, but I'd never chance it on hard disks.Why not? Don't you take backups?

[vengence]

Seriously, who's going to run 4 drives in RAID 0? Unless, of course, you like a really low MTBF. I don't know how SSD drives compare with rotating rust on reliability, but I'd never chance it on hard disks.

I think it largely depends on what you store. I don't have really anything on my disk that would cause drastic harm if I lost.

[player-x]

Seriously, who's going to run 4 drives in RAID 0? Unless, of course, you like a really low MTBF. I don't know how SSD drives compare with rotating rust on reliability, but I'd never chance it on hard disks.
I would also not run it in R0 but think lots of people would consider it for 2 drives
I'd also question calling the Highpoint 4xxx series entry level raid cards. Maybe for an enterpise system, but for home use, I'd consider the 2xxx series entry level. Especially with only RAID 0, where you don't have to calculate parity, the dedicated IOP on the 4xxx series doesn't do you a lot of good.
Performance wise i would consider all products from Promice and HighPoint entry level
Unlike your average integrated RAID, most dedicated RAID cards will support online capacity expansion, allowing you to add a drive to the array and have the array still be usable with the OS running while the card reconfigures the drives (which can take a while). The can also handle hot spares, email & audible alerts if a drive fails, but I suppose that doesn't matter if you're running RAID 0, you'll find out when nothing works.
I don't a lot of people would disagrea on that whit you i would also use R5 ore R10 if i would use more then 2 drives, whit more drives you ad, you increase failure rate expectational
While it may not be true for Intel chipsets, I've always found the drivers for NVIDIA SATA RAID to be really flakey, if you could get them to work at all. I've never had driver issues with dedicated disc controllers (both RAID & other).
The intel chipset raid options are 10x better the nVidia's still they lag miles behind compered to Adatec and ARECA controllers, at a price ;)
I have two systems with Highpoint 2xxx series cards (a 2320 & 2310) and couldn't be happier. I bought them solely for online capacity expansion and to avoid driver issues. I'd never use an integrated RAID controller for anything other than RAID 0 or 1 on a pair of drives.
Noting wrong whit those controllers, they just a little slower then Areca en Adaptec but like i sayed preformance comes at a price a high one

[Brahmzy]

This discussion is serously outdated. Welcome to 5 months ago.
Keep in mind these topics and more have been covered at length on the OCZ SSD forums, with far more accuracy. You should stop by there if you really want an education on the cutting edge.
Also, boot times vary completely per brand, model and even size of SSD, so take all of those benches with a grain of salt. This has been verified on the forums, and I have timed it myself, using multibranded R0 setups.
Another thing, while a high-dollar controller is hugely beneficial on a traditional disk RAID0 setup, the benefits are far less with SSD. The ICH10R has proven to be "the" sweetspot for most consumer SSD setups. nV's onboard offerings are still a joke in comparison.
I'm waiting for the '11R.
Posted via [H] Mobile Device

[The Hunter]

I would have liked to see some discussion of access times and random reads/writes, particularly how they scale from 1 drive to 4. Those are the areas in which SSDs really make a difference - given enough spindles, HDDs can post pretty high sequential speeds, but for the majority of users (even "power" users), it doesn't make that much of a difference - how often are you really transferring data of the scale that 800MB/s would make a huge difference? For a few things sure it would speed it up considerably, but not to the amount that all those random accesses would going from ~10ms to 0.1ms.
Kyle et al, if possible it would be great to see an addendum with small random write/read speeds, access times and maybe IOPS for 1 to 4 disk setups. I suspect as with boot times, we'd see the performance decrease as the array got larger, but I don't know, and it would be great to see some conclusive figures.

[mryerse]

Hmmm, well I think that games and program load times could benefit from the high transfer rate, although it seems mostly the OS will not need this as much, and certainly media would, although I doubt many will use SSD to store their HD video collections.

Personally I'd be really concerned about failure in a 3-4 disk striped array and would seriously consider raid 5, in which case I wonder about the performance of ICH10R vs an Areca loaded with mem.

[The Hunter]

Hmmm, well I think that games and program load times could benefit from the high transfer rate, although it seems mostly the OS will not need this as much, and certainly media would, although I doubt many will use SSD to store their HD video collections.
That's certainly true, although I don't know enough about the internals of modern games to know how they load levels - I suspect that while there are definitely some pretty big texture files, for every one of those, there are probably dozens of smaller files for which the random access speed is more important.

When you get down to it, everyone's usage is going to be different, which is why it's helpful to have all the info you can get, which is why I'm very curious to see the non-sequential test numbers - there have been a few of these SSD RAID tests around, but they all focus on the sequential speeds, and Anandtech (otherwise the king of SSD testing IMO) hasn't really bothered with any RAID tests.

[interested]

SSD are definitely the future when it comes to containing the OS and apps.

One problem right now is that vista and windows 7 ( 7 is supposed to be SSD optimized) both take up way to much space for the OS.
The problem lies in "C:\Windows\winsxs"
This amount of space this folder occupies grows over time, and gets to the point that it can easily eat up half the available space on a 32GB SSD.
(winsxs = 12.6GB on one of my Vista64 drives)
Note that this is even with a Vlite install, and with system restore off.

It would be nice if Microsoft made a utility that allowed people to regain some of this space, to make using small SSD drives more acceptable as an OS drive.

For more info read the following link that was posted on the windows 7 forums.

http://social.answers.microsoft.com/Forums/en-US/GettingReadyforWindows7/thread/450e0396-6ba6-4078-8ca0-b16bf4e22ccf/

[mryerse]

One problem right now is that vista and windows 7 ( 7 is supposed to be SSD optimized) both take up way to much space for the OS.


Isn't that the truth... turning off auto recovery helps a lot (making sure you're still well backed up provided you even need it for an OS/prog partition), as well as disabling hibernation (hibernate.sys is huge), as well as disabling the page file provided you have enough ram. This is an interesting argument for disabling the page file as in the past the argument has been "let Windows manage it for you, don't mess with it..." and I typically wouldn't argue that but in this case if MS insist on making their OS take up gobs of disk then I'd be disabling page file on a small SSD for sure.

[interested]

Isn't that the truth... turning off auto recovery helps a lot (making sure you're still well backed up provided you even need it for an OS/prog partition), as well as disabling hibernation (hibernate.sys is huge), as well as disabling the page file provided you have enough ram. This is an interesting argument for disabling the page file as in the past the argument has been "let Windows manage it for you, don't mess with it..." and I typically wouldn't argue that but in this case if MS insist on making their OS take up gobs of disk then I'd be disabling page file on a small SSD for sure.

Yup I have done all those things when installing vista or win7 on my small SSD, its almost essential on a small SSD to do those things in order to free up space.
And using vLite also helps slim down installation size.

As i mentioned though, right now even after doing all those things, there is still no "safe" way to free up the large amount of space occupied by the winsxs folder.

Also if you read up on what the winsxs folder's purpose is, you will find that it also is of questionable value to all users, thus it would be nice if Microsoft would provide a power tool to allow people to decide if they require it at all, or if they would prefer the old Xp non winsxs method instead.(see the windows 7 forum link i posted for more info)

It also sucks that this folder continues gets bigger over time, eventually consuming an even larger portion of precious SSD space.

So on one hand you will want to use windows 7 due to it being optimized for SSD's , but on the other hand you will want to use windows xp since it uses up far less space, thus actually allowing you to fit it on a small SSD.

[mryerse]

Yeah I went and read about it and it seems like it's for compatibility and that maybe it would be nice if you could just disable the feature provided you know ahead of time what programs you will use and if they'll work with windows 7 or not. Alternatively you could just get a bigger SSD. Maybe this is on purpose to boost SSD sales. I like conspiracy theories.

[veritas7]

I bet there will be more SSD optimization features by the time Win 7 reaches RTM/gold/retail.

Or at least by SP1... :\

[mikeblas]

You can't simply disable side-by-side; if you do, programs stop working.

Windows 7 has already gone gold. SP1 will be the first opportunity for fixes.

[INFINITE]

Was this tested on 32 or 64 bit Windows 7? I didn't seem to see that noted anywhere.

[extide]

ALL of windows is in that SxS folder, and the files you see everywhere else are just hardlinks. When you view the properties of the windows folder most of the files are getting counted twice, or more, in the total size count, so that can be kind of misleading. The disk free space and used space on the overall drive is correct though.

[tricky0502]

Was this tested on 32 or 64 bit Windows 7? I didn't seem to see that noted anywhere.

if it was tested on a 32bit os i think im going to stop reading this site!!!

you are right though, it doesnt even tell you what the test system is other than it is a x58 mb with ich10r.

[KOOLTIME]

To bad dint toss in a couple of WD 300 VRapts in raid for compare, so the real "HIT HOME RUN" or "NOT" would be seen as SSD vs HD's are for the x58.

WD's HD are about $200.00 each currently,
and the current SSD listed new egg has em for $660.00 each.

That rocket raid set up with those 4 drives would cost someone approx $3k ( taxes n shipping etc) retail.

Its to bad though, would be really sweet if a SSD maker could just make one with enough storage for the market at a HD type price they would sell a ton.

KT

[mikeblas]

Right. The lack of comparison to a known control set, the failure to do provide much explanation of the observed measurements, and the lack of experimentation (like changing controllers or stripe size) makes for a really weak article.

[extide]

I dont really know if comparing it to Raptors is even worth the effort, I mean seriously, does anyone think that Raptors are even remotely close to as fast? I agree that they should have tried a really small stripe size as SSD's might like that...

[Dekoth]

To bad dint toss in a couple of WD 300 VRapts in raid for compare, so the real "HIT HOME RUN" or "NOT" would be seen as SSD vs HD's are for the x58.

WD's HD are about $200.00 each currently,
and the current SSD listed new egg has em for $660.00 each.

That rocket raid set up with those 4 drives would cost someone approx $3k ( taxes n shipping etc) retail.

Its to bad though, would be really sweet if a SSD maker could just make one with enough storage for the market at a HD type price they would sell a ton.

KT

Right. The lack of comparison to a known control set, the failure to do provide much explanation of the observed measurements, and the lack of experimentation (like changing controllers or stripe size) makes for a really weak article.

I dont really know if comparing it to Raptors is even worth the effort, I mean seriously, does anyone think that Raptors are even remotely close to as fast? I agree that they should have tried a really small stripe size as SSD's might like that...

There are more then enough articles on a Raptor raids maximum speed to be able to easily compare speeds. Raptors aren't even in the same ballpark as an SSD anymore, raid or not.

Now Enterprise level 15krpm Cheetah drives, might be worth an argument, but then you are going to make the SSD's look cheap.

[mikeblas]

Now Enterprise level 15krpm Cheetah drives, might be worth an argument, but then you are going to make the SSD's look cheap.Hardly. SAS drives are between ten and 15 times cheper than SSDs.

[R0N1]

Nice testing, though a short one. Some results from other drives would also been interesting to see.
Even if I did know that it takes time for the RAID controller to load and initialize the drives when using RAID, I didn't expect the time to increase that much by every drive. Though the load time would still be almost fast enough. ;)

[mikeblas]

I dont really know if comparing it to Raptors is even worth the effort, I mean seriously, does anyone think that Raptors are even remotely close to as fast? I agree that they should have tried a really small stripe size as SSD's might like that...Raptors are far cheaper, though, even when RAIDed up. While the SSDs are probably faster, we don't know how much faster since we're given no frame of reference. Is the price difference worth the added performance?

There are more then enough articles on a Raptor raids maximum speed to be able to easily compare speeds. Using the same hardware and setup as this article? If so, then this article could have easily linked to the others as a reference. If not, then those other articles aren't useful as a baseline and this article should have established one.

[Dekoth]

Hardly. SAS drives are between ten and 15 times cheper than SSDs.

Granted it has been a long time since I purchased enterprise 15k scsi drives, so you may be right on that.

Raptors are far cheaper, though, even when RAIDed up. While the SSDs are probably faster, we don't know how much faster since we're given no frame of reference. Is the price difference worth the added performance?

Using the same hardware and setup as this article? If so, then this article could have easily linked to the others as a reference. If not, then those other articles aren't useful as a baseline and this article should have established one.

For the most part the other hardware is going to make little significant difference on the raw throughput differences. You might see some slight variances here and there and even possibly a few anomalies here and there, but they are going to be fairly easily compared. My point is a baseline against platter drives here was not important because it has already been done in past articles. Since hard drives are fairly predictable in this manner, there is no need to continually review the same set over and over.

[mikeblas]

Granted it has been a long time since I purchased enterprise 15k scsi drives, so you may be right on that.You don't even need 15K drives; 10K will do. And you don't need to purchase them; just look at the prices.

For the most part the other hardware is going to make little significant difference on the raw throughput differences.You've theorized that it should make little difference. But it may, or it may not. Actually running the test would validate your assumption, and help readers know if there's anything wrong with the other numbers that were presented. This is fundamental scientific method, and driven by critical thinking.

[extide]

Raptors are far cheaper, though, even when RAIDed up. While the SSDs are probably faster, we don't know how much faster since we're given no frame of reference. Is the price difference worth the added performance?


I used to run 4 160GB raptors in a raid 0 stripe. It was fast. Now I run ONE SSD. It is faster (and cheaper).

(The raptors DID cream the SSD in sequential reads, I was getting over 400MB a sec on that, but the actual usability and speed of the computer is faster with the SSD. I can get just under 250MB a sec with the single SSD in sequential reads.)

[mikeblas]

I used to run 4 160GB raptors in a raid 0 stripe. It was fast. Now I run ONE SSD. It is faster (and cheaper).What are the prices? Which parts did you buy?

The drives used in this test, Corsair P256s, give 256 gigs of storage for $659 (http://www.newegg.com/Product/Product.aspx?Item=N82E16820233085&Tpk=corsair%20p256). The 300 gig Velociraptor is $230 -- before rebate (http://www.newegg.com/Product/Product.aspx?Item=N82E16822136322). It's about a third the price and gives 20% more storage.

(The raptors DID cream the SSD in sequential reads,It was faster ... except when it wasn't, then?

[extide]

I meant the cost of four raptors vs one ssd. Yes, less storage space, but I dont care about that, I have a lot of network storage where I keep all my stuff.

It was faster ... except when it wasn't, then?
The raptors benchmark faster in SEQUENTIAL reads, which almost NEVER HAPPEN. In random access the SSD blows them away. I can open ALL of Office in about 2 seconds. That would take probably 20-30 on the four raptors. It's not even a fair comparison. My machine boost up faster with the SSD , shuts down faster with the SSD , I can open Visual Studio as fast as notepad. It's nearly impossible to make my computer actually lag now with the SSD, it doesnt matter what I am doing. Games load faster. Visual studio took ~15-20 minutes to install on the four raptors, it took 2-3 minutes on the SSD. Can I make it any clearer? I mean we are talking an order of magnitude faster in most of these comparisons...

[HecHacker1]

Interesting scaling, but what is your configuration?

I would love to see a comparison between:
Microsoft AHCI vs. Intel Matrix Driver
AHCI + Windows Software raid
Matrix + WIndows Software raid

The intel matrix driver makes a big difference, or at least used to compared to the default AHCI driver. I remember when the intel matrix driver had many bugs that caused pauses and stops, or crazy loading times. Most of that is fixed, but still how does it compare?

I don't see a lot of sites mention the difference or take the time to benchmark.

And is ATTO really reliable? I would rather have a simple Windows 7 file copy, in various sizes of mixed files/large files/config files. And how about a game install, and program install times (mixed file writes). And game loading times (mixed file reads).

Thanks for the hard work. I would like to see "real world" benchmark much like you do with your video card reviews.

[Dekoth]

You don't even need 15K drives; 10K will do. And you don't need to purchase them; just look at the prices.

You've theorized that it should make little difference. But it may, or it may not. Actually running the test would validate your assumption, and help readers know if there's anything wrong with the other numbers that were presented. This is fundamental scientific method, and driven by critical thinking.


You are arguing for the sake of arguing for a point that is irrelevant. The information you seek has been done and redone in dozens of articles out there concerning hdd speed. There was utterly no reason for them to include something that anyone who has done more then 10 minutes of casual reading should already know. That is akin to asking them to include a Geforce Ti4600 as the baseline in every modern video card comparison.

[mikeblas]

You are arguing for the sake of arguing for a point that is irrelevant.Feel free to find another conversation where you don't have to "argue"; and take your appeals to ridicule, hasty generalizations, special pleading with you. If you don't understand the relevance of scientific method in studying the performance of these products, it's not my problem.

[|mkanet|]

I'm surprised The Hunter is the only person who even brought up random read/writes on SSD RAID 0 as the array gets larger. I also suspect that performance degrades for this as well. If true, this would probably make a "real world" difference in every day stuff like booting windows, exist apps (writing configuration data to disk), loading apps, etc.

In fact, it would be nice to know in general who Windows every-day performance is affected when scaling RAID 0. If I'm paying that much for SSD disks, I sure as hell better see a noticable difference in common Windows tasks if I buy more than one for RAID 0. Currently I own one P256 and the Intel ICH. I wouldnt have it any other unless of course its significantly faster to get another disk for RAID 0.

I would have liked to see some discussion of access times and random reads/writes, particularly how they scale from 1 drive to 4. Those are the areas in which SSDs really make a difference - given enough spindles, HDDs can post pretty high sequential speeds, but for the majority of users (even "power" users), it doesn't make that much of a difference - how often are you really transferring data of the scale that 800MB/s would make a huge difference? For a few things sure it would speed it up considerably, but not to the amount that all those random accesses would going from ~10ms to 0.1ms.
Kyle et al, if possible it would be great to see an addendum with small random write/read speeds, access times and maybe IOPS for 1 to 4 disk setups. I suspect as with boot times, we'd see the performance decrease as the array got larger, but I don't know, and it would be great to see some conclusive figures.

[metalsmart]

The Windows start times are a curiosity. Is there any insight into this? Perhaps I missed it, but if the O/S shows a performance decrease on start-up with a RAID setup, how can I guarantee other applications won't be affected?

In reference to 2+ SSDs, can anyone attest to significant real-world improvements in boot-times, application start-up, and load times?

[Pyrolistical]

I'm surprised The Hunter is the only person who even brought up random read/writes on SSD RAID 0 as the array gets larger. I also suspect that performance degrades for this as well. If true, this would probably make a "real world" difference in every day stuff like booting windows, exist apps (writing configuration data to disk), loading apps, etc.

In fact, it would be nice to know in general who Windows every-day performance is affected when scaling RAID 0. If I'm paying that much for SSD disks, I sure as hell better see a noticable difference in common Windows tasks if I buy more than one for RAID 0. Currently I own one P256 and the Intel ICH. I wouldnt have it any other unless of course its significantly faster to get another disk for RAID 0.

From what I've seen random read/write on raids, latency takes a small hit but the bandwidth scales well. In fact its almost linear scaling.