Sign in to follow this  
TimmyHH

'Slight' SCSI RAID performance problem

Recommended Posts

Both logical drives are being used in the same software environment (Win2K SP3). The problem is not with SP3, though... I also tested right after installing the original Win2K release and got the same results.

MB: Asus A7V333 (VIA KT333A)

RAID config: Dell PERC2/DC with 2x Cheetah X15-36LP, RAID0, 64k stripe size, write back cache...

One of the Cheetahs alone should do 60 MB/s... why take the risk of RAID0 if reads aren't faster, and writes are much slower than in a single-drive configuration?!

I get the feeling that either the chipset's PCI implementation truly sucks (then why does IDE RAID not have this problem?), or the controller has some problem.

Here's the ATTO result for my WD800JB:

atto800jb.gif

And here's the result for my Cheetahs... :?

attox15.gif

Any suggestions? Maybe I should just get a decent U160 card and run them as single drives?

Share this post


Link to post
Share on other sites

Unfortunately, I don't have a 64-bit PCI slot anywhere near me... I would consider buying a MB that supports it, but only if I was sure it would fix the problem.

What I did try was an Intel BX-based MB... I thought maybe it was just a VIA problem again. But the performance on the BX was actually worse... the array managed about 40 MB/s read/write. (That machine was also running Win2K, BTW...)

The strange thing is that, although it wasn't as good as it should be, read performance under WinXP was much better than now!

atto-x15-raid0.gif

The only reason why I installed Win2K was stability... the XP drivers for this controller are somewhat poor, causing the array to fail randomly.

(The system would crash, and the management software would log things like "Physical drive 255:255 failed. Sense code 00, sense key 00, ...." etc.)

Share this post


Link to post
Share on other sites

OK, I think I'm going to get rid of this controller... I just powered up, and it hung detecting the drives. After about a minute, it complained: "Unresolved configuration mismatch between NVRAM and disk(s)"...

The drives are probably OK... I ran SeaTools on them, as well as a media verify... no problem found.

Now, if you where in my situation, what would you do?

- Use the X15s as single drives?

- Use them as software RAID0 - not feasible because you can't use it as a boot drive (correct me if I'm wrong)

- Make a RAID1 of two WD JBs and put the X15s somewhere else? Would that be faster than a single X15, in terms of STR?

Share this post


Link to post
Share on other sites

Hi.

Well, it has nothing do do with the PCI-bus since that caps out first at 133 Mb., and you are not near saturating it, considering your scores.

Second, what RAID-card did you say you used? you should for a configuration like that, go for the Adaptec 23320-R, which is a very good card ( 64 bit, but 32 bit compatible )for RAID 0 and has room to grow even further in capacity later on. If you spend so much money on two VERY GOOD disks you should choose a controller card that matches them. :o

This would set your scores right up where they belong! :lol:

/Yzerman

Share this post


Link to post
Share on other sites

The PERC2/DC is actually a dual-channel U2W AMI MegaRAID card. It also supports 64-bit PCI (as well as 32-bit), and unlike the Adaptec 29320-R, it's a true hardware-based RAID controller with an i960 CPU, 64 megs of cache...

So I doubt that the 29320 would, under normal circumstances, be faster.

But as the PERC2/DC seems to be incompatible with almost anything except Intel server MBs featuring 64-bit PCI, the Adaptec card with it's bootable software-RAID approach could actually be worth considering... has anyone actually used this card?

Share this post


Link to post
Share on other sites

Well now that i know what card you have it kind of confirms my suspicions. That kind of Hardwarebased, XOR processing RAID card is almost NEVER a good card to run RAID 0 with. Thay are often optimized for larger arrays and is always a very good RAID 5 performer.

/Yzerman

Share this post


Link to post
Share on other sites

Thanks for pointing me to the 29320-R, Yzerman.

Looks like I'll be ordering one tomorrow... it seems like a good compromise between (RAID 5-optimized) hardware RAID and non-bootable software RAID.

The only downside is that there are no Linux drivers that support RAID drives... but I figure it might be possible to install Win2K on the RAID0 array, and connect a third drive (to the same controller) to install Linux on?

Share this post


Link to post
Share on other sites

"Any suggestions? Maybe I should just get a decent U160 card and run them as single drives"

No, run them in Raid0. I get 80mb/sec read and 60 mb/sec write with my Adaptec 29160 and 2 older Quantum 10kll. So your 2 drives should be faster. But i use software Raid0 with dynamic disks.

François

Share this post


Link to post
Share on other sites

hi again.

Good thing that you are thinking of the 29320-R i use it and are more than happy with it´s performance with my two X15 n´s. :lol:

I´m just curious what you mean when you say there is no RAID-drivers for linux?

/Yzerman

Share this post


Link to post
Share on other sites
Good thing that you are thinking of the 29320-R i use it and are more than happy with it´s performance with my two X15 n´s. :lol: 

I've now ordered it... should arrive here soon, probably tomorrow. I'll report back on how it works for me when everything is set up...

I´m just curious what you mean when you say there is no RAID-drivers for linux?

At the Adaptec website, there's a list of supported OSes. Everything except Windows NT/2K/XP is listed with the appendix "(Non HostRAID)", so until someone perhaps develops an OSS Adaptec HostRAID driver, you're stuck in single-drive mode.

Share this post


Link to post
Share on other sites

I just installed the 29320 with the two X15s in RAID0. The read speeds have indeed improved by about 50% to 90 MB/s... however, I still get a maximum write speed of about 40 MB/s, which is obviously way too low.

Now I decided to put the card into my recording PC, which is an elderly PIII-700, but it's the only system I still have which is based on an Intel (BX) chipset. Guess what... the write speed went up to 70 MB/s. Perhaps not quite where they should be, but I could live with that...

The obvious solution now is to use a different (non-VIA) MB... what does everyone think of the AMD 761 chipset based MP boards? For example the Asus A7M266? Or should I go Intel straight away...? (Which I'd like to avoid because of the extra expense for CPU and possibly Rambus memory)

Share this post


Link to post
Share on other sites

Hi.

I´m glad to hear that you did benefit from your buy of a 29320! :lol:

However, as you mention the write speed is too low.

Do you run the array as a BASIC disk? If so this can count for some of the performance loss you are having. In that case convert to DYNAMIC disk at once.

I also want to say that I have an INTEL board ( with rambus, which, today is not alot more expensive than DDR memory ), and my scores are in the 100+ area.

Remember that in the case of a "not just a gaming machine" memory latency is important but SUSTAINED memory transfer rate even MORE important. This is where RAMBUS comes to it´s right!

plz keep me posted. We are going to get your array up to speed!

/Yzerman

Share this post


Link to post
Share on other sites
Hi.

Do you run the array as a BASIC disk? If so this can count for some of the performance loss you are having. In that case convert to DYNAMIC disk at once.

I did that before even running ATTO for the first time.

I also want to say that I have an INTEL board ( with rambus, which, today is not alot more expensive than DDR memory ), and my scores are in the 100+ area.

True, it's not much more expensive than DDR these days. The point is that I already have 512 megs of DDR333...

I've done some research in the meantime, and it looks like the AMD chipsets are not at all free of stability issues... albeit much better than VIA.

I may just bite the bullet and go for a P4 upgrade... this seems like the best long-term solution. Am I right thinking that the i850E would be the way to go for best PCI performance?

Share this post


Link to post
Share on other sites

Ok.

I´m running a regular 850 INTEL chipset-based motherboard from abit, and uses RDRAM PC 800.

Go for the 850 E with 1066 support. You won´t regret it.

You might want to know that SiS will launch a board with support for the new RDRAM PC1200... :lol:

/Yzerman

Share this post


Link to post
Share on other sites
I also want to say that I have an INTEL board ( with rambus, which, today is not alot more expensive than DDR memory ), and my scores are in the 100+ area.

True, it's not much more expensive than DDR these days. The point is that I already have 512 megs of DDR333...

I've done some research in the meantime, and it looks like the AMD chipsets are not at all free of stability issues... albeit much better than VIA.

I may just bite the bullet and go for a P4 upgrade... this seems like the best long-term solution. Am I right thinking that the i850E would be the way to go for best PCI performance?

Configured properly VIA chipsets give good results. My system can bench just over 120MB/s on my VIA Apollo Pro 266T chipset. Don't believe the crap that VIA chipsets don't perform. It's true that many of them need a form of latency patch installed but it's not a hard thing to do..

Share this post


Link to post
Share on other sites

I'm aware of the fact that most of these problems are probably fixable through a BIOS or driver update... but why doesn't VIA (or Microsoft or any other company) do it?

The main difference is that Intel chipsets work right out of the box, no monthlong patchwork required. If an Intel product does happen to be fundamentally flawed, they usually stand behind it and correct the problem - not really true of VIA...

Share this post


Link to post
Share on other sites

Have you tried the latency patches available for the via boards? Or, better yet, Powerstrip to adjust the latency of your scsi card? Adjusting the latency of the card on the bus can almost equate you read/write scores, or so I've seen on my own system and a few others. You might want to give it a shot. Definately the GB latency patch.

ff

Share this post


Link to post
Share on other sites
I'm aware of the fact that most of these problems are probably fixable through a BIOS or driver update... but why doesn't VIA (or Microsoft or any other company) do it?

The main difference is that Intel chipsets work right out of the box, no monthlong patchwork required. If an Intel product does happen to be fundamentally flawed, they usually stand behind it and correct the problem - not really true of VIA...

The BIOS is not VIA's fault, it's the mobo manufacturers'. I've read of plenty of cases where no patches are required, so obviously at least some of them are writing their BIOS correctly.

And VIA DO have a driver that corrects the problem for most people. Unfortunately they chose to name it poorly and called it the VIA RAID Performance Patch, when it doesn't only apply to RAID setups.

Share this post


Link to post
Share on other sites

OK, I've now reinstalled one more time and tested ATTO scores with the latest 4in1 drivers, the George Breese latency patch, and the VIA RAID performance patch installed. None of these made any noticeable difference - the results may have increased by about 5 MB/s, but this may just as well be normal variation.

I also tweaked with some BIOS settings on a KT133-based Epox board (my own Asus A7V333 doesn't even have these settings), causing the write speed to alternate between 30 and 50 MB/s, instead of 40 MB/s constantly... what an improvement :roll:

I've pretty much given up and started looking for a P4 board right now...

Share this post


Link to post
Share on other sites

I just installed into my machine an Asus P4T533 MB, a P4 2.53 as well as 256 megs of PC1066 RDRAM. All I can say is: this machine rocks. I expected it to be faster, of course, since the P4 replaced an Athlon XP 1900+ ... but I didn't think it would matter that much!

Oh, and as could be expected, my ATTO scores now seem right. I'm getting ~100 MB/s read and ~70 MB/s write.

Thanks go to everyone who replied to this thread...

Share this post


Link to post
Share on other sites

I have about 52 to 53 megs of read and write on my (VIA) Dual P3 MB, with a Cheetah 36ES 10krpm and an Adaptec 29160N on a partition positioned at about the center f the drive, so no VIA problem here, performance was the same using a i850 based MB and P4.

Share this post


Link to post
Share on other sites

If this is indeed the same old PCI latency problem VIA has been unable to fix for the last few years... then you don't have it because your single drive does just 50 MB/s.

Up to a certain transfer rate it works fine... but if you surpass that 'limit', you get a dramatic dip in performance, because the chipset gets 'overloaded' and puts in extra waitstates.

In other words... if my drives were just writing at 50 MB/s, I wouldn't have a problem. But because they manage 80 MB/s, the problem surfaces, causing performance to break down to 40 MB/s...

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
Sign in to follow this