Raid or not raid?

[Gurm]

I speak from experience on the CPU utilization.

To do a HUGE filecopy from my VIA bus (686A, Ultra66) to the Promise drive (FastTrack 100, Ultra100, 2 drives) takes around 25% of my CPU _AVERAGE_. The machine crawls while doing it.

HDTach reports that the average CPU utilization for straight read/writes on the RAID is about 8%. That's a lot, considering your average Ultra100 controller runs about 4%, maybe less.

- Gurm

[Paulr]

The Promise Fasttrak TX4 looks like a lovely card, 4 IDE channels for amazing performance.
I actually owned one for a week!!!
But, they aren't compatible with P4 motherboards at the moment
I'm now using a TX2 and it's great, I've used Highpoint based RAID controllers in the past, but it's true what they say about getting what you pay for.
The Promise controllers maybe a little more expensive, but they perform so much better, as I said previously a Promise controller running RAID 0+1 offers almost the same performance as a Promise controller running just RAID 0.
The only exception to that rule, the Adaptec is expensive and seems to offer rather lackluster performance.

[Dr Mordrid]

Gurm,

Please note that when copying from RAID to another controller/drive you are using TWO devices at once, not just the RAID. Therefore you can't lay off the whole thing to the RAID.

What portion of that percentage of that total is due to the slower drive and its controller/OS cache etc.?

How much of your CPU utilization is being caused by VIA's lousy busmastering support bogging things down?

Is HDTach using buffered transfers or is it SSW/R? If buffered then buffer I/O's and HDTach itself are using a few percentage points.

Is the FT's S.M.A.R.T. support on or off? If on then this adds overhead due to S.M.A.R.T. constantly polling the drives in the array.

etc. etc. etc.

Just asking....

Dr. Mordrid

[Gurm]

1. Don't know how to toggle SMART on the FastTrak (mine is a modified PD20265 built into the motherboard).

2. The other drive has 4% sustained average transfer overhead.

3. I'm basing the 25% on how much windows explorer says is being used.

4. I'm not saying that some latency twiddling can't fix the issue, just that it's not so miraculous as y'all might think.

- Gurm

[Dr Mordrid]

You turn off the S.M.A.R.T. function in the FastCheck utilies Options page. I presume FastCheck comes with all versions of the Fasttrak, embedded or not.

If you don't have it isntalled and running then do so. It monitors the array during use and will flag a failed array as it happens. Most often this is just a loose cable, but you never know...

As for how much CPU time is used by what; total CPU utilization isn't the whole story. There are lots of things going on besides what the hardware & its drivers are doing.

To show this I just finished running the Win2K Performance utility in Win2K during an RT-2000 HDBenchmark run. This was then set up to display the Total CPU utilization and those for the HDBenchmark and Fasttrak services.

Note that the blue line is the Fasttraks process and that it barely gets off the deck;



What's interesting is that the benchmark itself can take up ~50% of the Total CPU allocations. The remainder of the Total CPU then is likely Windows using its disk cache and other filesystem operations not directly related to the Fasttraks driver.

Dr. Mordrid

[Indiana]

Even with my (generally considered inferior to Promise) HighPoint RAID I really don't notice much slowdown due to the RAID even when copying large files (and I'm speaking of 10+ GB, here), not even when doing it from the RAID to my onboard VIA IDE controller. Not really that much difference from copying to my SCSI (it's measurable maybe (at least HDTach tells me so), but not noticeable).
Point is that when you need 30+ MB/sec SUSTAINED transfer you'll need one hell of an expensive LVD drive or a cheap IDE RAID. At least I've had not that much luck doing vidcaps when using Win2k's software RAID compared to the HPT370.

[Dr Mordrid]

Yup...and what's really nice is something like the FT TX4 that benches at 85-92 mbytes/s with a good set of 4 IDE's.

Talk about being capable of high transfer rates

Yes, it is useful...especially if you're doing realtime with a stack several effects high. Just using two DV streams and an effect requires about 13 mb/s at least.

Add more effects layers and the throughput requirement skyrockts fast, especially if the video is at higher than DV bitrates.

Dr. Mordrid

[rugger]

RAID, questions?

Where I work, we have a software raid-5 SCSI array (5 disks, 9gig each) running under linux. It seems to perform ok (enough for our needs, even in degraded mode).

Yesterday, one of our drives decided to die, but it did so by time-outing SCSI commands and causing SCSI bus resets, and not by giving bad sectors directly. Eventually, the hard drive system would stop responding to commands completely, gradually locking the system up. I had to manually remove the drive from the array and run fsck on the array before bringing the system back up (still much better than rebuilding and restoring from backups)

Anyway, my question is, would having hardware RAID have noticed the drive, properly disabled it, and allowed the array to coninue running without manual intervention?

[Dilitante1]

in a word

YES


-Dil

ps the reason why rugger is the hardware will notify you of the issue but not kill the operation of the PC/Server as it sees the array is still functional....Adaptec/AMI raid controllers act this way

[Dr Mordrid]

I gave up on pure software RAID's way back in my NT4 days after such a failure. Never again....

Cards like the ones discribed at least have monitoring software that will notify you of a drives failure and often present options for handling it. Many even offer hot-swapping with compatable bays & caddies and the abilitiy to rebuild the array.

Dr. Mordrid

[rugger]

Recommendations for SCSI RAID card

Ok then

Can someone recommend to me a RAID-5 capibable SCSI RAID card that has good performace/ good reliablity, but not too pricy

[Paulr]

RAID-5 SCSI Card.
Good Performance - That's not a difficult one, they are all very much of a muchness.
Good Reliability - I can certainly give you the names of some manufacturers of SCSI RAID cards who have sold me very reliable cards.
Not to pricy - And here is where we hit the slight problem, the words SCSI RAID 5 & Not too pricy I'm afraid really can't be used in the same sentance.
Mylex make a great SCSI RAID card, I've used quite a few of these (www.mylex.com for models).
Adaptec as always are quite good, but I've had two Adaptec cards swapped out for Mylex ones due to certain issues with the Adaptec models.

To get the most out of a RAID 5 array you are also going to want to look at hot-swapable chassis/caddies that way your server will not suffer any downtime should a HD fail, you simply remove the fault drive, insert the new one and all the RAID to re-build on the fly without taking the server down at all.

[Dilitante1]

rugger

I picked up a nice "used" AMIMegaRAID 428 card for $75 from
www.microexpress.com 2-3 months back

might be worth a look/call for you
make sure you get a stick of parity ram(72 pin) to park on it too =P



-pickle

edit:

http://www.teamexcess.com/

has them for $99 in stock now