"True" RAID-0 Performance

[unlockthesource]

I am running 2xWD 80GB special edition's in RAID-0. They are on a Promise card, on the PCI bus, each with its own 80-pin cable.

According to Sandra, my rig is quite a bit faster than your standard single drive, yet nowhere near the speed of a theoretical RAID-0. Compared to a non-RAID IBM Deskstar setup, it is usually 2 or 3 seconds faster on 512MB data copies, but the same on game loading.

Has anybody actually had any success pairing drives in RAID-0 so that you get optimal performance?
What stripe size did you use?
What controller?
Anyone with more than 2 drives in RAID-0? (very interested in this one)

 

[imported_nitrus]

1) 16k
2) onboard raid is best, if not promise is good but just make sure that the card has updated bios and drivers.
3) Raid 5 is best with multiple drives due to data integrity.

i honestly dont think raid is worth it due to cpu utilization, especially for your setup becuase it is running on the PCI bus. my two 74 raptors are setup as follows :
drive 1) os 2) pagefile, temp internet files, program (games, utilities).

overall its alot smoother.

 

[sao123]

The PCI bus will limit your speed somewhat, especially if you have a lot of other cards...
IE sound card, modem, Network Card...etc. Onboard raid is much better, because it stems directly from the south bridge instead of the pci bus.

SATA raid 0 is faster than a standard IDE ATA100 PCI raid.

 

[Cheetah8799]

I run RAID 0 with a Promise PCI controller and 2 Seagate 7200.7 2mb cache 80gb drives. I have noticed an increase in speed from my single WD800jb drive, but mostly with loading games, loading game levels, and things like that. Most applications load the same speed. Larger files get accessed faster.

My main reason for doing it was 1) to learn something, and 2) to improve gaming speeds, which is important to me with this rig. I have succeeded with both.


I chose Seagate drives due to noise concerns. WD and other brands are just too noisy for me. The 8mb cache Seagate drives are quite a bit more expensive than the 2mb ones as well...

The cost of RAID is up there, but so is dual CPUs, and dual anything... There is a reason people do it though, and if you want that extra bump up in speed, then you have to do it. Or get a 10k rpm Raptor... But be prepared for a lot of noise, and a hefty price for a single drive...

 

[unlockthesource]

Most of you say that the PCI bus is slower than the onboard. I would like to know how you came to that conclusion. Bus speed is measured in bandwidth, clock speed and congestion. I have only sound on my PCI bus. The clock speed of PCI os 33Mhz. What speed does the IDE/SATA bus transmit at on your mobos?

CPU utilization is not an issue, Promise controller manages the striping. It is transparent to the OS and CPU.

Note: i may have understaed my performance, the point is, it is nowhere near twice the speed of a single HD.

Note2: i also have onboard SATA RAID, but havent tried it

 

[sunase]

>According to Sandra, my rig is quite a bit faster than your standard
>single drive, yet nowhere near the speed of a theoretical RAID-0.

I haven't taken a look at Sandra (and if it has it's own marker for where theoretical RAID0 for similar drives to yours should be then nevermind ^^), but realize that synthetic benchmarks generally won't show a double score since they spread the IO operations around as opposed to doing a straight run test of transfer rates. I think this short changes the user experience effects a little since a lot of the time you are waiting on your hard drives you are waiting for a large file and smaller/multiple file access are too quick to have much impact on what you're doing anyway, but I do edit video so. ^^

>CPU utilization is not an issue, Promise controller manages the
>striping. It is transparent to the OS and CPU.

Sorry, not true. The controller *and the drivers for it* manage the striping. Most of the cheap consumer RAID cards (as in Promise/etc rather than Adaptec/etc) do very little in hardware and have the drivers handle most of it since it's cheaper. For example I was looking at Linux SATA RAID chips support the other day, check out the huge number (including Intel's ICH5R) that are marked "proprietary software raid" (and the author's recommendation to just use linux software raid since it's programmed better, hehe):
http://www.linuxmafia.com/faq/Hardware/sata.html

Indeed Promise doesn't do well against even Windows software RAID in the CPU usage department as per this article at ars (admittedly it's old and with slow drives, though):
http://arstechnica.com/reviews/4q99/fasttrak66/fasttrak66-2.html
Do remember when checking out CPU usage, however, that it's not the usage for RAID vs. 0% that you should be considering, but rather CPU usage for RAID vs. CPU usage for doing the operation without RAID.

>The clock speed of PCI os 33Mhz. What speed does the IDE/SATA bus transmit at on your mobos?

OK, PCI is a 32-bit bus,which is 4 bytes. So it theoretically tops out at 4*33.3=133.3 MB/sec (note these are "decimal" MB, like most numbers here, since most of that number is from the M in MHz ^^). The popular onboard ICH5R SATA RAID has 150MB/sec bandwidth for each drive, but only a 266MB/sec interconnect between north and south bridge.

WinBench's beginning transfer rate is one number where this is noticeable. On my machine a single 73GB Cheetah 15k.3 gets 76 MB/sec, again on my machine two in Adaptec RAID0 but on the PCI bus get 119. Now 2 WD740 Raptors on ICH5R get 143, however, which comfortable beats my drives. That last number is from this review:
http://www.extremetech.com/article2/0,1558,1541671,00.asp
I know StorageReview reports 76.4 for a single 73GB 15k.3 Cheetah, so at least for that number you know my testing is OK since they are close. ^^

So even with two drives and nothing else running on the PCI bus taking up it's shared bandwidth you can lose a little bit of performance. Your drives are slower, however, so it's less of an issue. Also serious RAID users need more than 2 drives so ICH5R isn't much help there. I'll link my WB transfer rate graphs below just for kicks. It's so sad seeing that array top out due to the bus limitation, isn't it (it stops early since that's when I took the screenshot ^^)?

http://pics.atofftopic.com/Images/wai/raid_rate.gif
http://pics.atofftopic.com/Images/wai/single_drive_rate.gif

edit: some cleanup, that's the longest post I'll ever do again...

 

[txxxx]

In short, RAID-0 on a desktop user configuration (e.g. gamers) will show little difference to justify 2x data vunerability. I'd switch them back to being single drives personally. Data > speed, especially with such little gain in a desktop environment.

 

[sunase]

>will show little difference to justify 2x data vunerability

Isn't this a performance thread? It's not like we don't have dozens of other threads to cover the other issues of RAID. ;p It's sad we have to have people attacking the reliability strawman here as well. Anyway, re reliability, personally it's been *years* since a drive died on me without enough warning for me to get the data off it. Even if one did, everything important I have is in folders that are constantly backup up anyway (with large amounts of disk space being dirt cheap nowadays such is easily accomplished).

If you are so scared of failure rate *you should be backing up anyway*, even with a single drive. So having to backup due to using RAID0 is no more effort. In fact, since the point of RAID0 is performance (can always JBOD if you want spanning with less risk and don't need the performance) you're likely to have smaller faster drives being put together rather than larger ones to go along with the RAID produced performance. This makes backing up even easier since the space in question is smaller. I can easily copy my entire array over to one of my storage drives when I feel like it, although incremental backups are better of course.

 

[txxxx]

Originally posted by: sunase
>will show little difference to justify 2x data vunerability

Isn't this a performance thread? It's not like we don't have dozens of other threads to cover the other issues of RAID. ;p It's sad we have to have people attacking the reliability strawman here as well. Anyway, re reliability, personally it's been *years* since a drive died on me without enough warning for me to get the data off it. Even if one did, everything important I have is in folders that are constantly backup up anyway (with large amounts of disk space being dirt cheap nowadays such is easily accomplished).

If you are so scared of failure rate *you should be backing up anyway*, even with a single drive. So having to backup due to using RAID0 is no more effort. In fact, since the point of RAID0 is performance (can always JBOD if you want spanning with less risk and don't need the performance) you're likely to have smaller faster drives being put together rather than larger ones to go along with the RAID produced performance. This makes backing up even easier since the space in question is smaller. I can easily copy my entire array over to one of my storage drives when I feel like it, although incremental backups are better of course.

Not everyone does. I know where I'd stick my money and config.

 

[unlockthesource]

Data vulnerability is an interesting topic txxxx.
True, i have doubled the odds of my data being destroyed. The reason i have decided to take that risk (and it came after MONTHS of consideration) is that i dont really feel any more at risk in RAID-0. Allow me to elaborate before the posts come pouring in.

• I have never had a drive die on me. My father had one, and S.M.A.R.T. started yelling at a Maxtor drive of mine a few years back. The maxtor was replaced, and i gave it to my sister as a spare ... its still operational.

• Being in non-raid will NOT spare you from data loss. Both of my hard drives were very important to me in the pre-raid days. The loss of either would have been no less devestating than the loss of both.

• All of my apps, music and movies are on CD/DVD. I have an extensive collection of TV rips that i would hate to lose, but i could replace those too.

Bottom line:
Your entire machine could be fried tomarrow by some freak power supply manfunction. Take a chance.

 

[unlockthesource]

sunase, you bring up some good points. In not sure how accurate the CPU utilization stuff you were talking about is for my card. Its not on the list you linked to, and i have found CPU Util. to be nominal during data access (<3%).

I didnt know the bus was so much faster for the Intel chip. One day ill switch to the onboard SATA raid and see if theres some difference.