Is RAID-0 REALLY faster???

[mcvan]

I am stumped about RAID 0 and its advantage(s). Despite extensive tweaking and fine-tuning, I can?t get my file system in W98SE running any faster than on my older slower non-RAID system. In fact, the new system often appears slower to find files and open folders than my old one!! This is despite various benchmarks (SiSoft Sandra, Norton, etc) that indicate the great superiority of the RAID setup (Drive Index=40,000).

Actually, all the RAID benchmarks are roughly twice as good, except for one: Average Access Time is 14 ms in RAID, and 8 ms in my old non-RAID system. Does this explain the difference? Also one, of my drives is setup as a slave. Does that hurt access time?

Although there are lots of posts about troubles setting up RAID and so on, I?ve seen very few about RAID performance. It would be great to hear of others? experience with this. I?d love to get a real discussion on the topic going here. Are there special BIOS settings I need to know about? Tweaks in Windows? Any enlightenment would be most welcome!

One other question for someone in the know: SiSoft Sandra says my AGP bus speed is 94 MHz and the PCI bus is 47 MHz. Is this for real? Those numbers seem too high for any of the components to survive ? isn?t the PCI standard 33? Or is Sandra misreporting the data?

The 2 systems:

AMD T-bird 850 @ 980 (140 x 7) w/GW FOP32-1 cooler
ABIT KT7A-RAID w/ ?YH? bios
512mb Infineon PC133 CAS3 RAM
Landmark 296 case / 340w PS, 1 front case fan
Matrox G400 Max
2 IBM 75GXP HDs - 46G, in raid-0 setup
Misc: SB Live, Plextor 12x cd burner, Tosh dvd, USR Sportster modem, ATA100 ZIP250 drive, Acer 10/100 NIC
Summary: Runs cool (27-43°C), perfectly stable. Not possible to go beyond 1G or 143mHz FSB. Either CPU or memory limit. System is 2 weeks old.

Intel P3-550 @ 780 (142 x 5.5) w/ CO-P302 cooler
MSI BXMaster w/ latest bios
512 mb Micron PC133 CAS3 RAM
Generic ATX case w/250w PS
Asus 7100 MX-32 twinview video card
1 IBM 75GXP HD ? 30G
1 Maxtor 30G HD ?7200 rpm (used only for backups)
Misc: SB Live, generic CD drive, USR Sportster modem, Acer 10/100 NIC
Summary: Runs cool (30-45°C), perfectly stable. System is 1 year old

 

[Noriaki]

<< Actually, all the RAID benchmarks are roughly twice as good, except for one: Average Access Time is 14 ms in RAID, and 8 ms in my old non-RAID system. Does this explain the difference? >>

There you hit the crux of it. In most people's normal use Access time is by far the most important important number. STR (Sequental Transfer Rate) is very unimportant compared to Seek Time.

Most hard drive benchmarks put too heavy an emphasis on STR over Seek...and in average usage STR just isn't an important statistic.

Not so say RAID0 is useless, if you are doing Video editing then STR is fairly important, but in &quot;average&quot; usage, STR is a nothing stat compared to Seek time.

Now having said all that....your seek times shouldn't be that much worse in RAID0 than a single drive. I don't think RAID0 helps performance significantly, but it shouldn't hurt it either. Might be a problem with your RAID card or one of your drives...

Edit: Oh yeah, also IDE RAID is (except in the case of a special few cards) is software based, so it sucks CPU time as well...probably not a major issue though, a TBird @ 980Mhz is pretty quick.

(BTW Seek Time and Access Time are interchangable terms if you didn't gather that yourself)

I know jack about Sissoft sorry...I care very little for synthetic benchmarks (and the RAID0 issue is as good as reason as any for my ignoring synthetics).

 

[Vinny N]

Noriaki hits it on the head though I wouldn't go around using access time and seek time interchangeably because access time pretty much a real number with rotational latency tacked on whereas seek time is the ideal that is often quoted by a drive manufacturer.

BTW, what block size did you use when you set up the array?
Small 16k ones or larger 64k ones? If I remember correctly, 64k ones should work better overall with less access time though it will benchmark in those &quot;flimsy&quot; artifical tests as slower than an array with 16k blocks.

As for what Sandra reports, maybe you should see if another program backs that up.

I would THINK that the KT7A uses a 1/4 PCI and 1/2 AGP divider for bus speeds of 36mhz and 70mhz respectively.

However, Sandra is either accurately reading the KT7A using a 1/3 PCI and 2/3 AGP divider, or it is calculating those numbers itself with those &quot;incorrect&quot; dividers. 1/3 PCI would result in 47mhz, and 2/3 AGP would result in 94mhz.

 

[mcvan]

Yeah, I suspected as much, that the access time difference is probably what I am seeing. How to get it better than 14 ms? Dismantle the RAID? (What a pain!) I have to admit, I love how quiet RAID mode is -- the IBM drives are pretty quiet anyway, but in RAID-0 I can barely hear them at all.

I think I chose 32k for the striping block size.

One tip I saw somewhere is that Win98/FAT32 performance can be improved by disabling synchronous buffer commits and/or write-behind caching for drives in the File System Properties. Any comments on this??

Chris, your thoughts about the way Sandra is reporting those bus speeds sounds about right. I'll see if I can't confirm that elsewhere.

Thanks for the quick posts!

 

[KK]

hey all...how much of a performance gain for video editing can one see from installing this raid capabilty? I am going to be installing a fasttrak 100 raid card with 3 30GB 7200/100 drives on a A7V MB with a 1.0Ghz@1.25 TB and 768MB Ram. Most of the video will just be bringing it in via a camcorder to a Radeon 64 VIVO and then converting it down to a usable size.

Thanks for any input.
KK

 

[mcvan]

Going through some of the other RAID-related threads, one comment I noticed was that the drives on a RAID-0 should be on separate channels for best performance. I had set up my drives on the same cable, as master and slave. I reset both drives as masters and hooked up another 80-pin IDE cable. Amazingly, the Highpoint controller identified the new setup as a bootable RAID and up came Windows, no harm done.

Access time has now improved to 9 ms, which is close enuf to the non-raid drive. I can actually see the difference using win explorer. So for the time being, I will keep the RAID. One day I'll have to play with a hardware raid controller to see whether there's much improvement.

Thanks for the insights, guys.

BTW, KK, it seems like your application is perfect for RAID to make a big performance boost, according to all the info out there...

 

[MGMorden]

The board can't read/write to both a master and a slave at the same time, that's why you were getting such bad scores w/ the RAID that way. Just in case you were wondering . . .

 

[RedShirt]

That was exactly what I was going to suggest. Also, to any other potential RAID buyers out there, IDE RAID above 2 drives really isn't worth it.

Go with 2 drives each a master, after that, the preformance increase is barely anything, too much data to go through those IDE cables.

 

[rmblam]

KK and whomever is interested

FYI, the A7V is almost RAID ready as is; in case you did not know. Crazyape has the info on an easy mod (no soldering) to a bit harder mod (soldering) to activate the onboard RAID.

More info is at also http://www.tweakhardware.com/raid/

I prefer the soldering method, especially since the easy mod has some Win2k issues. Mine does not. Crazyape and the guys at TH may have it resolved by now though.

also http://forums.anandtech.com/arcmessageview.cfm?catid=27&amp;threadid=323962&amp;highlight_key=y&amp;keyword1=my%20a7v%20raid%20mod

 

[mcvan]

Red Shirt: Have you -- or anyone else reading this -- tried any other IDE RAID devices? Are your comments based on real experience or theory? In other words, is the performance limitation intrinsic to IDE RAID or is it this &quot;software-based&quot; iteration of it (the Highpoint 370 RAID of my Abit MB) that's the limiting factor?

For example, Storage Review did a voluminous comparison between the Promise Super Track 100 and the Escalade 6400 IDE RAID cards. They thumb down the Promise card but conclude

<< The Escalade's performance should put to rest the false impression many folks carry about ATA RAID; that it provides no real-world performance increase. The Escalade's performance proves this notion false. >>


So I am wondering, how much real gain will I see using something like the Escalade 6400 in Win98 running DTP/graphic software, office apps and the odd games?

Bottom line is that I don't really give hoot about the numbers, I want a real improvement in my computing experience (unlike going from my P3-780 to this new T-bird 980. This was not a great move 'cus the gain ain't worth the effort or money. I should have gone to a 1.2G t-bird...)

 

[NOS440]

All I can add here is my experience with my IWill Side Raid 100 w\2 30 gig IBM 75's. I find it very responsive and has 9 ms access times tested with HD Tach. It opens large files in a snap and when playing Quake 3 online I am always the first one in a new level. It was never like this before I switched to the raid setup of coarse the P4 1.5 GHZ probably doesn't hurt either. But I had the same raid setup on my 933 P-3 and it was pretty much the same. Hope this helps

 

[sharkeeper]

I never thought there was much difference in everyday use between a single drive and two striped until I got out the stopwatch!

I was shocked especially at the reduced program launch times!

Cheers!

 

[Pariah]

&quot;Average Access Time is 14 ms in RAID, and 8 ms in my old non-RAID system&quot;

mcvan, what benchmark are you using? Whatever it is, it is highly inaccurate. There is no IDE drive that can achieve 8ms access time. The IBM drive you use should be in the high 12's, not 8.

&quot;Seek Time and Access Time are interchangable terms if you didn't gather that yourself&quot;

You will see this a lot, but it is false. Average seek time is the average time it takes the read head to find the correct track on the platter. Average access time is average seek time + average latency. Average latency is how long half a platter rotation takes which is directly related to RPM's.

&quot;I know jack about Sissoft sorry&quot;

All you need to know is it is one of the worst HD benchmarks available.

&quot;performance can be improved by disabling synchronous buffer commits and/or write-behind caching for drives in the File System Properties&quot;

I've never seen the synchronous buffer option anywhere, but write-behind caching will slightly, though probably imperceptively improve performance. This feature waits as long as it can for the disc to be free before it attempts to write the data in cache to the drive. The downside to this is, is if your computer crashes, any data pooled in the cache is lost before it can be written to the drive.

&quot;Most of the video will just be bringing it in via a camcorder to a Radeon 64 VIVO and then converting it down to a usable size.&quot;

If you are using a DV camcorder, the RAID array will be completely useless for bringing data off the camcorder, as the bandwidth required is only 3.6MB/s. Even analog captures can be handled by any dedicated current generation 5400RPM IDE drive today. Depending on how you resize the video (codec used) the CPU will more likely be the bottleneck instead of the array. RAID comes in handy when you do a lot of editing and manipulation with the video, adding filters, effects and what not.

&quot;Access time has now improved to 9 ms, which is close enuf to the non-raid drive&quot;

Once again mcvan, there is something wrong with your benchmark. Neither the RAID array or the single drive should come anywhere near that. Putting the drives on seperate channels will improve performance. Changing the drive placements won't affect access time, only STR.

&quot;Go with 2 drives each a master, after that, the preformance increase is barely anything, too much data to go through those IDE cables.&quot;

If you are talking about software RAID, or the cheap RAID cards, this will be true. Hardware RAID cards like the 3ware cards will see practically a linear increase in STR as you add drives as every drive has it's own dedicated channel.

&quot;Bottom line is that I don't really give hoot about the numbers, I want a real improvement in my computing experience&quot;

IDE RAID will not do much for everyday tasks. If you don't care about benchmarks, don't bother with IDE RAID. If you aren't on a budget there are other options that will improve the majority of everyday tasks.

 

[mcvan]

<< IDE RAID will not do much for everyday tasks. If you don't care about benchmarks, don't bother with IDE RAID. If you aren't on a budget there are other options that will improve the majority of everyday tasks. >>

Like what? I've got a 1G cpu, 512 mb PC133 RAM (been playing with virtual memory on/off), a great 2D card (G400 Max), and a tweaked up MB, what else is there? SCSI controller, 15k rpm drives &amp; RAID? A 1.5G T-bird? (P4 doesn't looks smart to me.) I think I am best off waiting for now, but let's hear about these other cost-no-object options.

 

[Pariah]

The true cost is no obstacle option would be solid state. Performance should be unmatched by anything available. The more traditional cost is no object would be high RPM SCSI are even high RPM SCSI RAID.

 

[zzzz]

sandra 2000 has a bug with the Kt133(a) chipset, newer sandra should correct it..
(for the agp and pci speeds)

 

[mcvan]

<< mcvan, what benchmark are you using? Whatever it is, it is highly inaccurate. There is no IDE drive that can achieve 8ms access time. The IBM drive you use should be in the high 12's, not 8. >>

Sisoft Sandra. I just found and ran this HD Tac mentioned here:

30G IBM HD: 11.3ms random access, 30 mbps read burst, 9.4% CPU util
IBM RAID-0: 8.6ms random access, 72mbps read burst, 28% CPU util

If these seem inaccurate, what do you recommend?