3 Velociraptors in RAID 0!

GriFF3n

Weaksauce
Joined
May 1, 2007
Messages
72
Just took the plung and bought 2 more velociraptors (300 GB) to add to my system. Going to have 3 in a RAID 0. Been looking online to see some benchmarks of comparable systems and have seen no buddy rocking 3 of these guys in RAID 0. Anyone here have any experience?
 
What is your primary objective? More space or more performance?

If you want performance, i would wait a bit perhaps and buy new third generation SSDs coming out in Christmas, which should be cheaper than currently and also faster, making it a good moment to join in as it were.

If you feel SSDs are too expensive then yes you can extend RAID0 (though not expand unless u use Intel onboard RAID). The added performance won't be that significant in overall performance, generally. But it's free performance so why not.

I do hope you have a means of backup and any data on RAID0 would fail if either drive fails. That also counts for single disks without RAID, of course. But multiple drives would amplify that risk. As long as you have a decent backup solution (like external 2TB 5400rpm drive) you should be fine.
 
Have had a RAID 0 fail on me before so I backup EVERYTHING on them now. Looking mostly for performance. Haven't made the jump to SSD's cause they lack capacity and are still a little expensive. I got these 2 vraptors from a forum member here for $170 shipped, so I naturally want to through them in the system as my OS with my existing vraptor to build a RAID 0. Was just wondering if anyone had and performance benchmarks to see how these bad boys will perform.
 
Those depend entirely on your setup; particularly which RAID drivers you use. You cannot compare Intel onboard RAID with AMD or nVidia onboard RAID, for example; those use different drivers.

With onboard RAID, all RAID is done by drivers; thus this is 100% software RAID mechanism. The difference lies in boot ROM support that allow it to boot from a software RAID volume, as Windows has no support for this itself. The quality of onboard RAID varies, but Intel is regarded as best 'fakeRAID' driver; and the only one which reaches acceptable write performance on RAID5 arrays.

The performance difference of another Velociraptor will be quite low as the increased IOps is still very low. You still have the latency of a HDD. I think it will be better to separate your system disk and your gaming/apps disk. For example, you could consider the following setups:

Setup A)
system/boot disk: New Velociraptor
Data/Gaming disk: RAID0 of 2x Velociraptors (the ones you already have)

Setup B)
system/boot disk: Small SSD to fit apps + OS but not games
Data/Gaming disk: RAID0 of 2x Velociraptors to store games

In other words, even if you don't opt for an SSD, having another velociraptor for system disk and leaving the RAID0 free for only one task, would be a better option than having only one array of RAID0 3x Velociraptor.
 
Got the drives in a here are the results;

zjy05z.png


I'm a little surprised about how the drive's performance doesn't decrease like the single vraptor. This is on an Intel chipset (Gigabyte ep45-ud3p). The drives are formatted with a 279 GB and 558 GB
sized partitions.
 
I spent $50 each and got 2x Western Digital WD5000AAKS put em in raid 0. Seems like a better bang for buck.

HDTune_Benchmark_Intel___Raid_0_Vol.png
 
If you use RAID you should not use HDTune or HDTach to test performance; you can't. Instead, use real benchmarks like CrystalDiskMark, AS SSD and IOMeter (more advanced) to test performance in realistic conditions; meaning: on the filesystem itself and not raw I/O directly circumventing the filesystem like HDTune and HDTach do.

Let's see how your RAID0 scales; it should scale close to 100% especially for sequential transfers.
 
If you use RAID you should not use HDTune or HDTach to test performance; you can't. Instead, use real benchmarks like CrystalDiskMark, AS SSD and IOMeter (more advanced) to test performance in realistic conditions; meaning: on the filesystem itself and not raw I/O directly circumventing the filesystem like HDTune and HDTach do.

Let's see how your RAID0 scales; it should scale close to 100% especially for sequential transfers.

I did't realize HD Tune wasn't good for RAID systems. I'll give CrystalDiskMark a try when I get home from work to see how she performs. I'm also looking up how to short stroke these drives to get a little bit more out of them. Thanks for all the advice so far sub.mesa.
 
Well the HDTune Pro version also has File Benchmark which does test on the filesystem. Still CrystalDiskMark and AS SSD also tell you about random I/O which is very important. Just sequential MB/s is only part of the story.

Also note for Intel onboard RAID engines you can enable 'write caching' which in HDTach and HDTune will cause the interface bandwidth/burst speed to go over 2000MB/s; because it is served from your RAM memory instead.
 
Here is CrystalDiskMark's output. Not really sure how it compares to others though.

21jrjwj.png
 
Those speeds look very reasonable to me!

Let me explain the numbers:

seq = Sequential and means transferring large files which are stored in one 'chunk'; i.e. the HDD only has to find this file once (1 seek) and then can read/write at full speeds.

512k = simulates file access of half a megabyte large; quite common but speeds on HDDs will be lower due to the HDDs having to seek every 512 kilobytes. Thus harder task for HDDs.

4k = small file access with only a single queue; things like booting and launching applications would include this kind of access. If you ever saw applications 'freeze' and then 'unfreeze' as the HDD finished the I/O then likely it's this kind of access. 4 kilobytes is very small and very hard for the HDD so under 1MB/s performance; which is very normal for HDDs. These speeds are most important when using the HDD as system disk (OS + applications + swapping).

4k QD32/64: this is the same as above, but with multiple queue depth. This means: multiple I/O requests at once; or rather 32 at once. This allows RAIDs to scale in random reads; you can see the 3.7MB/s being alot higher than 0.7MB/s; at least a good part of this increase should come from the RAID0.

Writing on 4K is different and depends on HDD firmware; good firmware would keep the 32 queue write speeds at same level as normal 4K.

End conclusion: those speeds look fine to me; it's what you can expect from your setup.
 
End conclusion: those speeds look fine to me; it's what you can expect from your setup.

Thanks for all that info sub.mesa. I'm still going to try short stroking and doing a fresh install of Win7. It's been a little over a year now so I think its time for a reinstall. I'll report back with some numbers once everything is back up to snuff.
 
3x Velociraptors in RAID 0 shortstroked
2qv4jvb.png


2x Velociraptors in RAID 0 Shortstroked
2l8xnio.png


1 Velociraptor (Originally had)
awyuyo.png


So looking to sell the other two vraptors on eBay and purchase a faster SSD. I did get some good gains, but not as much as I was hoping. Still a good learning experience though.
 
Back
Top