UPDATES!!!! (Single Drive Vs. RAID-0)

[chookman]

Hey Guys,

Ive seen alot of posts about people thinking about making a new system with RAID-0 but are unsure of the advantages/disadvantages. After my latest build (pieced together from some other parts as well as new ones bought) i did a small bit of testing of a single drive vs a RAID-0 with 2 of the same drives. If there is any other suggested benchmarks that you would like me to run ill give them a go, as i have another 500gb AAKS drive i can install for testing, i have only used HD-tune.

System:
P180
e7200 (OCed to 3.04ghz, i will be pushing this further soon)
Arctic Freezer 7 Pro
Asus P5KR mobo
2x WD 500gb AAKS
Asus 7900GT

First i used the single 500gb drive to install WinXP with only drivers installed (partition was full size of drive)

Second, i setup the RAID-0 in the configuration manager (Ctrl+H on boot) took about 2 seconds to do. I then installed Server2008 Standard (sorry for OS change but this is the OS that i will be using with this machine). Again partition was full size of array (~930gb)

As you can see there is quite a performance increase in Transfer rate (close to x2 the performance), access times stay the same (to be expected) burst rate about the same (also to be expected)... What i was a little perplexed with was the CPU usage, although this is more than likely just an anommoly, and i would expect the CPU usage to be lower than the RAID-0 array test. I didnt notice this until after i installed the RAID-0 drive otherwise i would have ran the test again.

EDIT: Just realised i didnt put in some disadvantages of RAID-0, The main disadvantage is that if one disk fails in the array you will loose all data on the whole array. RAID-0 is sometimes refered to the non-RAID, RAID of the bunch (or similar). This is because RAID stands for Redundant Array of Independant Disks, as RAID-0 isnt redundant its kind of a non-event. The other problem that some may encounter is more to do with creating an array with an onboard controller, this is because if the motherboard fails, or you simply wish to move the array to another machine it more than likely wont work. This is because there is no standard for the array information to be stored within the controller. This enables manufacturers so "make" there own system of information store, which other controllers wont recognise. So to transplant an array you will generally need exactly the same or similar chip/manufacturer card for the array to function once moved.

Just a quick bit of extra info for you, I also have a file server running RAID-5 on an Adaptec 3805 controller with 8 WD 500gb AAKS drives attached (3.18TB total) I ran a couple of tests on it as well, i was a little dissapointed with the Transfer rate, however, it is to be noted that the transfer rate doesnt drop over the sustained testing of the entire 3.18Tb volume which is more impressive. I also noticed that HD-Tune didnt seem to want to use the whole array for testing (2199Gb displayed) so i used HD-Tach to confirm results (which reported correct size), this must be some issue with HD-Tune and GPT disks. HD-Tach reported a slightly higher average transfer rate (~20mb/s).

The RAID-0 seems to not only boot into Server2008 very quick, but it also seems snapper in and around the OS itself. Might be interesting to test some games on loading times and possibly FPS figures as well, although im unsure how they will run on Server2008. So all in all im happy with the increase, considering the extra drive was only $95AU to purchase.

Hope ive helped some of you out there.

 

[Slobogob]

Nothing i didn't know already but a very nice idea indeed. Now let's hope that the noobs bother to even look for it instead of starting all kind of threads about it.

 

[chookman]

I think the newbies were the ones i was more aiming this at, thanks Slobogob

 

[4745454b]

If people look at this, they aren't getting the whole picture. This is a synthetic app, which would show AID0 to be a great idea. If you were to look at real programs, you wouldn't see a doubling of anything. (or a reduction in time by half...)

The problem I have with AID0 is people (noobs really) who buy a second harddrive or a soundcard at the expense of things that matter to gamers more. Why get an 8600GT if you are spending another $90 on a second harddrive, and another $50 for the soundcard? That $140 alone could get you an 8800GT. Add in the $60ish for the 8600GT, and you could get a nice 8800GTS 512MB, a much better gaming card. The AID0 setup will decrease the load times for sure, but the visuals with the 8600GT won't be as nice.

Server 2008 is based off the XP core is it not? (might be vista now that I think about it...) Try testing with some games or normal programs, not synthetic junk. I wouldn't mind seeing some more benchies.

 

[geofelt]

The only valid tests would be with the workload that YOU will be running. To the extent that your workload is a single task reading your files from beginning to end, then the hdtune or hdtach benchmarks are good indicators of your actual performance. Unfortunately everyone is different, and usage patterns can be unpredictable. www.storagereview.com has some tests with various configurations and workloads. They can explain why many are disappointed with raid-0 as a performance option.

A test I would like to see is how much difference a bit of overclocking makes at the 3.0 ghz level for games.

 

[halcyon]

AID-0 is fine as long as you're backing up your important data to at least one other independent disk (it is unlikely that one would lose the AID-0 array and the backup disk simultanously). No matter what you're doing, 2 Raptors (as old as they are) in an AID-0 array is fast...at least as fast as one and typically faster. So if speed is the concern an AID-0 array may be the remedy.

Just back up your data, because, even though I've yet to see it happen in the past 3 years I've read and I'm told AID-0 arrays do die due to disk failures.

 

[chookman]

Although im aware these are only synthetic tests, i mainly wanted a quick snap shot of the performance increase that myabe present.

I would be curious to know what kind of "tests" are run to get the figures in alot of review sites out there.

4745454b: Although im not using this machine for gaming ill give it a go, probably Crysis will be a tester, but also have to note its only got my old 7900gt in it. Server2008 is based of Vista, i had a few issues installing drivers but for the most part the vista drivers worked a treat.

geofelt: Can you explain more of what you mean by tests at 3.0ghz? How is this in relation to RAID-0?

halcyon: This machine is so far going to be a replacement (initially) for my DNS and DHCP server. Then as time goes by im going to use it with VMWare so im hoping the extra speed on the harddrives will help with this. Backup wise all my important stuff goes on my RAID-5 which is redundant and i will be looking at a "backup" process for some of this data as soon as i get some more dough.

 

[Slobogob]

AID-0 is fine as long as you're backing up your important data to at least one other independent disk (it is unlikely that one would lose the AID-0 array and the backup disk simultanously). No matter what you're doing, 2 Raptors (as old as they are) in an AID-0 array is fast...at least as fast as one and typically faster. So if speed is the concern an AID-0 array may be the remedy.

Just back up your data, because, even though I've yet to see it happen in the past 3 years I've read and I'm told AID-0 arrays do die due to disk failures.

That's why it's called Kamikaze-Raid.

 

[halcyon]

Even though I do backups I know one day one of my arrays will die. ...and it'll suck...and it'll take me some time to correct.

Until then, I'm enjoying the performance.

 

[4745454b]

If you have a 7900GT, you might want to look at something other then Crysis. Farcry, HL2, TF2, etc might make better games to test with.

 

[chookman]

4745454b: I should be able to give FarCry and HL2 a shot .

 

[SirCrono]

I've just runned the same tests in the same two disks and got very different results, I've got about 120 MB/s Avg on the raid array but the transfer rates were constant during the whole test, max at 140 and min at 94.
I blame the low speed to the RAID controller it's a no-name POS, but the speed decay in your test puzzled me.

Also my old WD2500JS beated the crap out of the raid in the HD tach burst test, 140 to 190 aprox (to lazy to re run the tests).

Any ideas why does this happens?

And by the way, great topic for claryfing RAID to all of us who aren't über knowledgeable. Thanks.

 

[PsyKhiqZero]

SirCrono
I would venture out and say that cpu overhead could have been the reason your burst test was lower on the raid. CPU needed time to think.

How full where was the raid array? HDD performance degrades as the disk gets full. At near full capacity my raid 0 got 170burst, after i cleaned up half way i got 220.

Not an expert and these are just educated guesses

 

[russki]

One thing I would like to see is the start-up benches (single vs. RAID) before you suggest that RAID is better in that respect. Paricularly with the change in OS, which I know you disclosed, but that essentially throws startup comparison and "feeling snappy" our the window.. In general, I am not a big fan of "feels certain way" anecdotal reports.

 

[SirCrono]

SirCrono
I would venture out and say that cpu overhead could have been the reason your burst test was lower on the raid. CPU needed time to think.

How full where was the raid array? HDD performance degrades as the disk gets full. At near full capacity my raid 0 got 170burst, after i cleaned up half way i got 220.

Not an expert and these are just educated guesses

I don't thick CPU overheating is an issue, the CPU was idling and it only gets to 40ºC on orthos and 42 on TAT.

The Array is completely blank, 0 bytes used, just installed and formatted.

 

[chookman]

One thing I would like to see is the start-up benches (single vs. RAID) before you suggest that RAID is better in that respect. Paricularly with the change in OS, which I know you disclosed, but that essentially throws startup comparison and "feeling snappy" our the window.. In general, I am not a big fan of "feels certain way" anecdotal reports.

Ill look at doing that also, im going to install Server08 on both single and RAID and check times for boot.

 

[4745454b]

While performance between these should be close (XP and Vista have semi close gaming performance now.) I would think load times would still be different. I'm not sure I'd put much weight behind anything that shows how long it takes different OS's to load.

Uhmmm, he didn't say overheating, he said overhead. I'm not sure thats the case though.

 

[geofelt]

OS startup is certainly an operation that is worthy of measurement to assess the impact of raid-0. It is repeatable, and something we may do every day.
One possible complication is the use of Vista ready boost and it's impact from learned behavior. Over time, ready boost is supposed to cache frequently used small files that will perform better from fast access vs fast data transfer.
I regularly use suspend instead of rebooting, and I suspect that raid-0 would be a big help there. Perhaps wecould get a few tests of that activity. On suspend /resume, there seems to be a lot of hard drive activity, probably saving and restoring memory contents. I recently installed the new WD velociraptor, and I notice an improvement in both boot and suspend/resume times.

Keep up the good work.

 

[halcyon]

OS startup is certainly an operation that is worthy of measurement to assess the impact of raid-0. It is repeatable, and something we may do every day.
One possible complication is the use of Vista ready boost and it's impact from learned behavior. Over time, ready boost is supposed to cache frequently used small files that will perform better from fast access vs fast data transfer.
I regularly use suspend instead of rebooting, and I suspect that raid-0 would be a big help there. Perhaps wecould get a few tests of that activity. On suspend /resume, there seems to be a lot of hard drive activity, probably saving and restoring memory contents. I recently installed the new WD velociraptor, and I notice an improvement in both boot and suspend/resume times.

Keep up the good work.

I think you mean Super Fetch...not Ready Boost (Ready Boost allows the use of removable memory (ie, Thumb Drives) to boost system performance.

 

[geofelt]

@halcyon:
No, I meant readyboost. Superfetch preloads stuff in anticipation of use, knowing your usage patterns. It happens after boot time.
Ready boost caches small files to your fast thumb drive for faster access later. Vista knows weather or not it has been removed and if the contents are still valid. If the contents are still valid, I think it could use them at boot time.

 

[halcyon]

@halcyon:
No, I meant readyboost. Superfetch preloads stuff in anticipation of use, knowing your usage patterns. It happens after boot time.
Ready boost caches small files to your fast thumb drive for faster access later. Vista knows weather or not it has been removed and if the contents are still valid. If the contents are still valid, I think it could use them at boot time.

Cool, thanks for setting me straight.

 

[4745454b]

It might be something we "may" do everyday (seriously? Who reboots every day?) but I'm sure there are differences between the loading of Vista and XP based machines. Unless he installs the OS, with the same level of patching, on both sets of harddrives, I fail to see how this test would prove anything.

Let me try to give you an example. We know that setting up an AID0 array doesn't cut load times for games in half. In fact, you might see only a few seconds difference. This is because there is more then just data transferring going on, there is CPU work involved also. (has to decompress the data, transfer data to and from the video card, etc.) I'm sure the same is true with loading an OS. Data would be pulled from the harddrive, then the CPU would have to decide what to do with it. Vista, with its caching features, might take a "long" time to load, if it has to put these small files in its various cache locations before it boots.

 

[chookman]

Hey Guys,

Just an update for some of you, I have installed WinXP Pro and SP3 on a single 500gb AAKS drive and the same on RAID-0 array.

First graph shows boot times for Windows off both setups, all times are in seconds

I dont understand why its slower with the RAID, the only thing i can think of is that i have done something different between the 2 builds.

2nd Graph shows boot time and load times for Crysis. This is v1.2 of crysis, and the loaded game was exactly the same (as soon as i got to shore at the start of the game i saved it)

We see here that the RAID-0 array seems quicker although only by a few seconds in each case.

3rd graph shows boot time and load times for C&C3. This is version 1.09 and the loaded game was an 8 player skirmish game with 7 brutal teams in.

We see here very little difference if at all. Averages are pretty much the same.

4th graph shows the results of copying a 2gb file from the root drive to the same location (effectively creating a duplicate file in the root directory) this should test copy speeds within windows.

We see here that times are significantly less on the RAID-0 disk, showing that RAID-0 helps with raw file transfers.

All up im pretty pleased with the RAID-0 performance (although OS load times were poor, could be from less than identical configuration on each disk). My main use for this machine will be DHCP, DNS and VMware machines, from these tests i should see better performance in my implementation. Although i dont think id recommend it for gaming as the results wouldnt really warrant the extra $$$ required. Hope this answered a few of your questions earlier, and if theres anything else let me know i should be fairly quick with results this time round.

 

[4745454b]

With regards to the load times of windows, I already said I wouldn't bother worrying about it. Your XP and Sever 2008 (vista based) have different cores, so they will/should have different load times. Seeing as Vista loads a lot of extra stuff into ram, I'm not surprised to see that it has the longer boot time. I wouldn't worry about improper configs, they simply have different things to load.

With respect to games, again, I'm not surprised. This only goes to show that what I've been saying for years now is true. If you are building a gaming rig, worry about the Raptors/AID0 for later. They only shave off a few seconds at best. Seeing as a second modern drive is $60-$100, that is much better spent on a bigger video card, or a better monitor.

The real "value" of the AID0 array is in file read/writes. Your final avg graph shows the array with a time of ~34(?)seconds vs ~52seconds for the single drive. (an 18 Second difference.) If you move files around, AID0 is worth it. If you are building a gaming rig, worry about it after you have a nice 22" LCD and a CPU/GPU combo good enough to run it.

Job well done btw, but where are the DVDshrink charts? (LOL, the job is never done.) Well done again sir, good job.

 

[geofelt]

@chookman. Keep up the good work.
Raid-0 shines with sequential processing. OS boot will load lots of small files, so seeking(arm positioning) comes into play. With a file split onto two drives, the positioning has to be done to both drives which adds to the time, and makes the faster data transfer not as important. I think you did the test correctly.

Could you repeat the file transfer test with the source file on one drive, and the destination file on the other drive? I'm betting that arrangement will be much faster than the raid-0 option.