Does Raid 0 actually improve performance?

[aznmist]

I used to think it does but after this article, it's giving me second thoughts.

http://faq.storagereview.com/tiki-index.php?page=SingleDriveVsRaid0

 

[Gurck]

I don't have any actual experience with it, although it would make sense since seek times depend on the drive(s) used and not how they're used. Still, for the performance enthusiast, a 10% performance improvement is no small thing.

 

[Barnaby W. Füi]

Actually, I can remember it being said in an anandtech article a while back that 10% difference in performance is the very borderline of being noticeable, so based on that, I would say yes, it is quite a small thing.

Raid 0 will give you better max throughput, which honestly probably isn't worth the trouble unless you have a specific need for lots of disk throughput. Only large-ish amounts of data shuffling will benefit from it, and your risk of data loss is greater.

 

[KristopherKubicki]

No. it does not.

Kristopher.

 

[snidy1]

Depends what your doing with your system.

 

[tallman45]

Just about any review I have seen addresses a comparison of a single drive vs a dual drive Raid 0. There is always one comparison that is omitted and in all fairness is difficult to measure, but none the less would definately be beneficial to most users.

Instead of 2 drives in a Raid 0, run 2 drives each on their own independant channels and balance you data between the 2. I have always found that files such as the pagedata and scratch disk on separate drives from the os and apps has offered substantial performance gains.

A Raid 0 array is great as an addition to your os/apps as a subsystem designed for video processing and digital processing of very large RAW photo's.

 

[xenos500]

Its a good bit faster. I had a an array on my old Abit KG-7 RAID using two IBM 60GXPs. Window XP booted up in about half the time it normally took. The XP loading screen bar would appear, move half the length down the display, and be done booting and at the log on window in a few seconds later.
No other computer, even faster newer ones I have made, booted as fast as that one. I dont know how much faith anyone puts in sisoft sandra anymore, but I ran it on both 60GXPS individually and they were capable of 8.9ms seek and 44MB/s max. As an array they scored 7.1ms seek and 82MB/s.

Im sure results vary depending on the drives and controller used. The abit motherboard I had used a Highpoint raid controller.

I never let the concept of increased chance of data loss bug me. I figured that if one of the drives broke, there was a good chance I'd have been using that one by its self anyways....and still lost all my data. To combat the possibility, I kept important things on my home fileserver.

I woud suggest to anyone to give raid 0 a shot.

 

[Regs]

Originally posted by: KristopherKubicki
No. it does not.

Kristopher.

Can you tell that to Tommy boy from the "other site" that? He thinks RAID is well worth it and 400 dollars is cheap.

The board has been flooded lately with RAID arguments because of it.

 

[KristopherKubicki]

Here is my opinion:

When do you use raid? I transfer a lot of several hundred MB files to and from some of my file servers; lots of 3D work. Here is a typical printout from an FTP transfer to ONE drive, no RAID, over gigabit Ethernet.

150 Opening ASCII mode data connection for directory listing.
226 Transfer complete.
TYPE I
200 Type set to I.
PASV
227 Entering Passive Mode (192,168,1,100,192,10)
RETR dummy_file
150 Opening BINARY mode data connection for dummy_file (331100448 bytes).
226 Transfer complete.
Transferred: dummy_file 315.76 MB in 13.30 (24,316.78 KBps)
Transfer queue completed
Transferred 1 file totaling 315.76 MB in 13.39 (24,316.79 KBps)

Now again off a RAID 0 stripe:
200 Type set to I.
PASV
227 Entering Passive Mode (192,168,1,100,192,19)
RETR dummy_file
150 Opening BINARY mode data connection for dummy_file (331100448 bytes).
226 Transfer complete.
Transferred: dummy_file 315.76 MB in 12.17 (26,564.27 KBps)
Transfer queue completed
Transferred 1 file totaling 315.76 MB in 12.27 (26,564.27 KBps)

Yes a 0.5 second difference to transfer a 300MB file. This isnt the most conclusive benchmark in the world but i just wanted to show people my experiences with the technology. Personally, I do not like doubling my odds that one of those drives will fail just so i get that 0.5sec advantage. I also don't really like the extra cost or setup. It all depends if that theoretical best case scenario of 4% speed increase is worth it to you or not.

Kristopher

 

[nageov3t]

unless you're running a database server, you're not going to notice the difference in speed by using RAID0.

IMO, the only RAIDs worth using are RAID1 and RAID10.

 

[gsellis]

Originally posted by: loki8481
unless you're running a database server, you're not going to notice the difference in speed by using RAID0.

IMO, the only RAIDs worth using are RAID1 and RAID10.

No, really, you do get a difference doing video editing. While editing, you have multiple digital video files open (avi or dif) and you have multiple rendered files of the dv files open. You never have enough memory to load all of that (remember - DV is about 13GB/hr). So, when you play in real-time, you need the bandwidth. Just wait until HDV which is around 19MB/s data rate. I guess you could call that a "database server" though. I already have plans for 500GB on my new system.

 

[oldfart]

remember - DV is about 13GB/hr

Correct. Any half decent IDE drive can do 10X that. I do a lot of DV video work (convert to DVD and WMV) and have never had a HD be any bottleneck in the capture/edit/encode/author process.

 

[gsellis]

Originally posted by: oldfart

remember - DV is about 13GB/hr

Correct. Any half decent IDE drive can do 10X that. I do a lot of DV video work (convert to DVD and WMV) and have never had a HD be any bottleneck in the capture/edit/encode/author process.

I still have issues when working on big projects (1.5 hrs) with lots of layers, even with RAID0. I may be an extreme though. The last big project was 110GB of DV files (from 20 tapes). The section that gave me the most I/O grief was merging 6 different angles.

It was worse with a single drive. Never had a capture/encode/author issue on it either. Just edit. I need more mem too, but the next machine will handle that. The next machine will be bigger (duals with a Gig) and I probably will put the Render area on a different drive pair.

I guess part of the answer would also be, if you really are constrained, RAID 0 is not a cure either, it helps with pain management.

 

[Amused]

I notice a huge boost in how "snappy" a computer feels in everyday HDD intensive tasks when running a RAID-0 array. From boot time to loading programs to HDD transfers to page file seeks, it makes things much snappier. It is also a MUST if you want to realize the full potential of a gigabit network.

The HDD is the largest bottleneck on a system. It makes sense to take advantage of any way to boost performance there.

I have a sneaky feeling the people saying RAID-0 is worthless have either never run a RAID-0 array, or didn't set theirs up properly.

As for safety, I have run various RAID-0 arrays since 1999 in all my machines and have had exactly 0 failures.

I'm currently running two Rapters in RAID-0 and this is, by far, the fastest non-SCSI system I have used.

 

[oldfart]

Originally posted by: gsellis

Originally posted by: oldfart

remember - DV is about 13GB/hr

Correct. Any half decent IDE drive can do 10X that. I do a lot of DV video work (convert to DVD and WMV) and have never had a HD be any bottleneck in the capture/edit/encode/author process.

I still have issues when working on big projects (1.5 hrs) with lots of layers, even with RAID0. I may be an extreme though. The last big project was 110GB of DV files (from 20 tapes). The section that gave me the most I/O grief was merging 6 different angles.

It was worse with a single drive. Never had a capture/encode/author issue on it either. Just edit. I need more mem too, but the next machine will handle that. The next machine will be bigger (duals with a Gig) and I probably will put the Render area on a different drive pair.

I guess part of the answer would also be, if you really are constrained, RAID 0 is not a cure either, it helps with pain management.

In your case, I could see it. It really depends on what you are doing. I see people automatically recommending a RAID or SCSI setup whenever video work is mentioned. In extreme cases like what you are doing, it may very well be a good idea. In most cases, its is not needed. I might work with 3 capture files at a time at most. Any decent drive is fine and not a bottleneck for that type of work.

 

[nny]

so most people agree that RAID 0 is a relatively small performance increase. instead, spend the money and get a Raptor 10k, which cuts the access time of most SATA drives in half to 5.2ms. my question is, if you already have a 36.7 raptor and need more HD space, would it be a bad idea to RAID 2 raptors together, or would that decrease the performance?

 

[beatle]

I have a 36.7 GB Raptor. When I want more space, I'll buy the 74 GB Raptor, not another 36.7 for RAID 0.

Remember, I/O heavy disk loads are best handled by SCSI, not ide, even RAID 0. Seek times will still be the same (I doubt any significant boost) for RAID 0 as it will be for a single drive.

 

[Markfw]

I have quite a bit of experience with raid0 in both IDE and SCSI. In both cases, transfer rates are limited to the bandwidth of the devices, for IDE 133 mps and in my case for SCSI (even though I have an ultra 160 controller) 133 mps since it in in a 32bit pci slot.

Where the real difference comes in, and it is considerable depending on the numbers of drives, is retrieving many different files, like booting the OS, or loading a game level. For these things you can see UP TO x times performance, where x is the number of drives in your raid array.

 

[Matthias99]

I don't care how many times people say this:

Seek times will still be the same (I doubt any significant boost) for RAID 0 as it will be for a single drive.

it's still not true. The MAIN benefit of a RAID controller (other than essentially doubling your read/write transfer rates when dealing with large files) is the reduction in effective seek time for heavy file I/O. If your data is split across N disks, it takes on average only 1/N times as long for a read head to get into the right position to read it. The effect is more dramatic with a RAID1 or 0+1 (where there are many disks that can fetch a particular piece of information), but even on a RAID0 it's there. While it's true that RAID cannot make your physical drives any faster (ie, reducing the minimum seek times for the disks), it can and does improve average seek times in real-world situations.

 

[GoSharks]

Originally posted by: Matthias99
I don't care how many times people say this:

Seek times will still be the same (I doubt any significant boost) for RAID 0 as it will be for a single drive.

it's still not true. The MAIN benefit of a RAID controller (other than essentially doubling your read/write transfer rates when dealing with large files) is the reduction in effective seek time for heavy file I/O. If your data is split across N disks, it takes on average only 1/N times as long for a read head to get into the right position to read it. The effect is more dramatic with a RAID1 or 0+1 (where there are many disks that can fetch a particular piece of information), but even on a RAID0 it's there. While it's true that RAID cannot make your physical drives any faster (ie, reducing the minimum seek times for the disks), it can and does improve average seek times in real-world situations.

uh i disagree. lets say that you have an ideal raid0 array where corresponding bits of data are stored in indentical, corresponding areas of two different hard drives. lets then create an imaginary situation where you are accessing data from the outside of the drive to where you are wanting to get data from the very inside track of the drive.

now assuming that your two drives have the ever so slight seek time difference, when one head has read the data that it has been assigned to read, the controller still has to wait for the other head to reach the corresponding data on the other drive. keep in mind that raid 0 splits your data into two pieces (or however many drives you put into the raid0), one place on each drive. you CANNOT do anything with only half the data that you need - you have to wait for the other half to come in.

 

[beatle]

Originally posted by: Matthias99
I don't care how many times people say this:

Seek times will still be the same (I doubt any significant boost) for RAID 0 as it will be for a single drive.

it's still not true. The MAIN benefit of a RAID controller (other than essentially doubling your read/write transfer rates when dealing with large files) is the reduction in effective seek time for heavy file I/O. If your data is split across N disks, it takes on average only 1/N times as long for a read head to get into the right position to read it. The effect is more dramatic with a RAID1 or 0+1 (where there are many disks that can fetch a particular piece of information), but even on a RAID0 it's there. While it's true that RAID cannot make your physical drives any faster (ie, reducing the minimum seek times for the disks), it can and does improve average seek times in real-world situations.

Matthias, we'd gone so long without butting heads over RAID 0 opinions!

Like GOSHARKS said, it can help improve rotational latency a touch, but only if both drives happen to line up perfectly over the data. My personal experience with RAID 0 vs single drive (same model, same identical hardware, just one drive vs. two) was that there was tangible difference. "Sampson Simpson, I stick by my story!"

 

[Pariah]

Originally posted by: Matthias99
I don't care how many times people say this:

Seek times will still be the same (I doubt any significant boost) for RAID 0 as it will be for a single drive.

it's still not true. The MAIN benefit of a RAID controller (other than essentially doubling your read/write transfer rates when dealing with large files) is the reduction in effective seek time for heavy file I/O. If your data is split across N disks, it takes on average only 1/N times as long for a read head to get into the right position to read it. The effect is more dramatic with a RAID1 or 0+1 (where there are many disks that can fetch a particular piece of information), but even on a RAID0 it's there. While it's true that RAID cannot make your physical drives any faster (ie, reducing the minimum seek times for the disks), it can and does improve average seek times in real-world situations.

That's only true if you increase the array capacity by the capacity of a single drive every time you add a drive which would be extremely wasteful for most people. If you limit the array capacity to the same capacity of the single drive, no matter how many drives you add, you aren't going to gain any positional seek advantage.

For example. If dealing with 40GB/platter drives, if you have one 5 platter 200GB drive, it will have to search the exact same distance on the platters as 5 single platter 40GB drives in RAID 0, meaning you gain nothing. In order for your calculations to work, you would have to add an additional 200GB drive to array every time. So in order to get 1/5 seek distance you would have to raid 5 drives for a 1TB array capacity which would be a complete waste if all you really need is 200GB.

KristopherKubicki, I don't understand the purpose of that test. When the ethernet connection is the bottleneck, what is your test supposed to prove? That RAID can't make your network connection faster?

As always, RAID benefits vary from user to user. For a typical home computer, RAID will provide pretty much no performance improvements in most scenarios.

 

[KristopherKubicki]

My gigabit ethernet wasnt the bottleneck.

Kristopher

 

[gsellis]

Originally posted by: KristopherKubicki
My gigabit ethernet wasnt the bottleneck.

Kristopher

Are you sure? Maximum theoretical speed is ~125MB/s. Since it is ethernet, figure a bit better than 50% of that actually. Seems to me that some drive combos can exceed that.

 

[dejacky]

If we're talking 8mb cache 7200rpm ata drives, then you're better off putting one on each ide channel and running them separately, the general consensus is that this setup is more useful especially for things like audio and video editing.

-dejacky