Think your HD is fast?

[Geeky1]

Beat this. I ran ATTO @ w/the default settings (except changing the maximum length to 32mb) on my dually system... stripe size is the default 64k...

Results:

[Spinner]

Very impressive indeed! What cluster size are you using 64 as well? What tweaks did you perform?

On a good day, and after quite a bit of tweaking, I can get mine nearing the mid nighties in reads, but your writes are well on their way past anything I've ever managed to get out of PATA's.

Nice work man.:thumbsup:

[Geeky1]

Everything's @ the default settings... no tweaks or anything (honest!)

Stripe size is 64k
The array itself is 2 DiamondMax 9 160GB drives in RAID 0 w/RocketHead adapters.

I just did a clean install of Windows (first install), loaded up the drivers for the chipset, video, etc., and ran ATTO.

[Spinner]

What's with the SATA convertor thing? could you explain that a little bit for me please.

I presume it's so you can run your IDE drives on a SATA controller correct?

[Geeky1]

http://www.highpoint-tech.com/rh.htm

They come with the HighPoint 1540 & 1520 if you buy the right version. You plug one end into the PATA drive's IDE connector, and the other end connects to a floppy power cable and a SATA cable. It simply converts Parallel ATA drives to run on Serial ATA buses...

[Al_Capown]

mine :/

[Spinner]

AL'

That must be one of the most bizarre ATTO screenshots I've ever seen. Try it with a total length of 32MB. Then post back.

Geeky1, I see, yea I had heard something about those adapters, but haven't had the pleasure of any first hand experience with them. Your bench results could be highlighting the inherent quality of the SATA controller, over that of a PATA one. Have you ever benched those drives when they were being piped through standard PATA controller?

[Geeky1]

Only individually. I don't have a PATA RAID controller, so...

By the way, I went with the RocketRaid 1540 because supposedly Promise sucks, and the RocketRaid 400 (440?) is only a 32 bit card... The 1540 is actually a PATA controller with 4 Marvel SATA converter chips on it. So the data is actually going from PATA -> SATA -> PATA -> RAID card. But whatever... it's fast. :D

[Al_Capown]

eh

[danball1976]

Yeah, thats how it should look like.

[t1rhino]

Forget the label b/c I miss labeled it.

[TheBaron]

t1rhino = t3h 0wn4r

[Geeky1]

T1... what are the specs on your drives? I switched from SCSI to the IDE RAID because my SCSI drives couldn't break 30MB/s in ATTO... they were 10,000rpm IBM Ultrastar LVD-160/SE drives w/4mb cache; both 36.7 & 18.2GB... and they just wouldn't do more than about 27000k/second...

[TheLostSwede]

The difference here is the scsi controller. I guess Geeky had a stinky Adaptec and Rhino has a LSI megaraid which hosts a Intel chip at 200 or 300 mhz with 128 mb of ram in cache.
It´s another league, not even compareable.
Also, the LSI controller retails around 1000 to 1500 bucks alone, but can be caught at ebay for around 100.

Great score Rhino!! Awesome!

Here´s the Raptors:

http://www.mackanz.com/attoraptors.jpg

[Spinner]

Al_Capown said
eh

I presuming that is just one single drive not in RAID correct?

t1rhino said
Forget the label b/c I miss labeled it.

:respect: Brilliant scores, fan-bloody-tastic! I guess though, you get what you pay for.;)

[danball1976]

Ok, on the left is my more normal score, on the right is if I disable Direct I/O on my Maxtor 120GB, the scores are the same drive.

[TheLostSwede]

Just change the left to overlapped and run it.

[danball1976]

Ok, same drive (120GB Maxtor). When did everybody start using Overlapped I/O on ATTO? It used to be where niether was used.

Oh, and my computer has been up for 6 days, 4 hours, so that affects the scores some.

[Geeky1]

Mack... great. NOW ya tell me. YOU were SUPPOSED to tell me BEFORE I bought the SCSI card... nevermind the fact that that was before I was @ Icrontic... you should've told me anyhow! :D ;D

Besides, what's wrong with Adaptec? I thought their 29160 was supposed to be excellent...

[TheLostSwede]

Basically, Adaptec has never released any card for the server market, more just for us normal beings. They have made a fortune on their name more or less.

http://cgi.ebay.com/ws/eBayISAPI.dll?ViewItem&item=2743495412&category=3753

Thats the card to get, i said wrong with he mhz before, dont know what i was thinking of. You can get theese cards for 100 bucks from time to time. Couple it with a pair of 10 or 15K drives which btw also goes for almost no money at all. Throw them into your dually on the 64 bit pci and you have a 250-300K atto.

Adaptec has always been toys where this is the real deal.
Try to get an Adaptec card to run OK on XP...fat chance. Ok, XP has very poor support for raid scsi setup, but anyway.

[Geeky1]

Hmm... how did they get their reputation then?
Any suggestions for non-RAID scsi cards? (I have 2 drives that aren't identical...)

The 29160 actually runs fine on XP, altho it's not a SCSI RAID card...

[TheLostSwede]

If the drives are the same size, you might get good speed if you softraid the drives in either 2K or XP. Just cause the cards have raid possibilitys doesn´t mean you have to run the drives in raid. Therefore, if i where you and if i where in the market for a scsi board, the 1600 would be my choice, hands down. But dont just rush and buy it, wait for a 100 bucks deal. You can always upgrade the drives later. If you have a good drive, you´ll see 100k atto on a single drive.

[Tex]

Mackanz said
Basically, Adaptec has never released any card for the server market, more just for us normal beings. They have made a fortune on their name more or less.

http://cgi.ebay.com/ws/eBayISAPI.dll?ViewItem&item=2743495412&category=3753

Thats the card to get, i said wrong with he mhz before, dont know what i was thinking of. You can get theese cards for 100 bucks from time to time. Couple it with a pair of 10 or 15K drives which btw also goes for almost no money at all. Throw them into your dually on the 64 bit pci and you have a 250-300K atto.

Adaptec has always been toys where this is the real deal.
Try to get an Adaptec card to run OK on XP...fat chance. Ok, XP has very poor support for raid scsi setup, but anyway.

pardon me? Uhhh.. XP has poor support for scsi raid ? I think you got that wrong. Xp does as well as any other OS with scsi raid really. Its a matter of the controller more then anything not the OS or drives.

And rhino's score above was not with the LSI megaraid controllers but with an even more expensive compaq controller available on their servers. Thats the one he posted at DF beofre he puilled from a new server at work. And on ebay they are more like 400+ dollars. I lust for one but just can't afford it....

This ATTO below is a fully loaded production system and the atto was pulled with almost a dozen things running and in the tray including MBM5 and Coolmon running with 5 second interval updates on them. This is a fully loaded system just as I use it everyday with indexing enabled etc.. and not a bare tweaked system trying to maximize the atto score. Ohh yeah.. I forgot to mention .....Its XP also... (grin)

Tex

[Tex]

Geeky1 said
Hmm... how did they get their reputation then?
Any suggestions for non-RAID scsi cards? (I have 2 drives that aren't identical...)

The 29160 actually runs fine on XP, altho it's not a SCSI RAID card...

You can get the same raid controller I posted bel;ow for maybe 50 bucks more then a nice used scsi card on eBay. I nice dual channel u160 oin eBay is 70 to 80 bucks. My raid card is more like 125. I run several single non-raided scsi drives off it.

here is an atto for one single scsi drive.

Tex

[Spinner]

Hey Tex, thanks for popping in, always a pleasure to share a thread with ya;).

All these SCSI bench results are making me depressed. There was me quite happy with my little old pair of basic PATA WD1200JB's, looking forward to my next upgrade, which will probably consist of a couple of SATA Raptors, and then you guys come long with scores getting on for double of what I can ever hope to get.

My bro had a straight up SCSI system (1 independant drive, running on a pretty standard controller), one of the old cheetah drives I think. The system they were in was brand new, he had just picked up a new PII 333. Almost the fastest desktop chip available at the time, and I still remember to this day, how fast that machine ran. It proved to me that, it doesn't matter what CPU you're packing, or what speed your system bus is running, if you've got a slow disk interface, then it all means nothing.

SCSI is obviously still the way to go for the best access times, but it's just so bloody expensive, especially when you want large capacity disks.

I envy those scores guys, I envy that you have the ability to be that close to a machine which packs so much pace. But I can honestly say, SCSI's price, is way too high for me. That's just thinking about the drives themselves, the high quality controllers you can get for them doesn't even bare thinking about.

Now stop showing off people, you're distressing your Uncle Spinner.;) :D

[Tex]

Spinner the controller and drives ran me a little over 350. thats for like 110gb of storage. The drives run 40 to 45 bucks a pop for 18gb. I have paid as little as 37 bucks and as much as 49 for a drive.

For the price of two new raptors...... Ahh never mind.... lets not even go there huh?

Mine also hit 118,000 on sandra BYPASSING the cache. The access time was 2ms. Check the ide raid access time when bypassing the cache.

Tex

[Geeky1]

The access time was 2ms

:wow: :wow: :wow2:

Dunno what the IDEs do w/o the cache, but with it, Sandra came back with a 6ms access time...

[Tex]

But with the cache its caching your reads and writes in ram from your system not hitting the disks/controller direct. It will be more like 9 to 13ms usually with ide raid.

Tex

[Geeky1]

yea... I'd expect it'd be more like 9-13ms since the drives are rated for like 9...

[Spinner]

Tex said
Spinner the controller and drives ran me a little over 350. thats for like 110gb of storage. The drives run 40 to 45 bucks a pop for 18gb. I have paid as little as 37 bucks and as much as 49 for a drive.

For the price of two new raptors...... Ahh never mind.... lets not even go there huh?

Mine also hit 118,000 on sandra BYPASSING the cache. The access time was 2ms. Check the ide raid access time when bypassing the cache.

Tex

I never really thought about purchasing second hand, I guess with HD's that's always been a no go area for me by default. But, like you said, you seem to have got yourself pretty well kitted out by obtaining your gear that way. I guess, if the mighty Tex is willing to do it Ebay, then so am I. :D

[Tex]

Between me and others I have helped we have bought 27 drives and had two to return. Thats not perfect but not bad either.

The biggest kill is the controller. They list for like over a grand with teh battery backed ram and stuff. Expect to pay about $125 if you take your time. Remeber on a non dualie system with a 32bit/33mhx bus your capped around 140,000 on atto give or take a little. With a 64bit/66mhx bus you can hit in excess of 500,000. With pci-x in theory you can maybe hit 1,000,000.

Thats in theory. Real life is a little diferant.

Here is the very best atto I have ever seen come off a 32bit/33mhz bus. Its a hpt 4 channle on a nforce2 with four old ibm 60gxp drives.

Tex

[MediaMan]

Thanks to my buddy Tex I have joined the SCSI elite with two x 18 GB ATLAS 10K III RAID 0 and a 9gb for page and temp files....all on a MegaRaid 1600 controller with 128 MB of RAM.

Did I get that right Tex?

Anyway...it's XP and after my first few weeks with SCSI...I have to say the main thing I notice is how those "sharp corners" of the OS are rounded off. Things just open and close "smoother" and many apps can run a little easier. It may not be 4000 percent faster...but I like it.

:)

Now to figure out what to tweak where to bump those scores up.

[Geeky1]

MM... your atto scores in SCSI are still much better than mine were :rolleyes: Trust me tho... IDE or SCSI, when you're pushing 100mb/s+ (or even 60-70mb/s+), it's noticeably faster than a single drive system... my biggest complaint with SCSI and RAID is the stupid boot times...

[Tex]

MediaMan said
Thanks to my buddy Tex I have joined the SCSI elite with two x 18 GB ATLAS 10K III RAID 0 and a 9gb for page and temp files....all on a MegaRaid 1600 controller with 128 MB of RAM.

Did I get that right Tex?

Anyway...it's XP and after my first few weeks with SCSI...I have to say the main thing I notice is how those "sharp corners" of the OS are rounded off. Things just open and close "smoother" and many apps can run a little easier. It may not be 4000 percent faster...but I like it.

:)

Now to figure out what to tweak where to bump those scores up.

Smooth is the way I try to describe the differance with dually's and scsi in general really. I think its the best way I have come up with to describe the feeling anyway.

Tex

[MediaMan]

Geeky1 said
my biggest complaint with SCSI and RAID is the stupid boot times...

I have to admit that the boot time is longer. I have to wait for the SCSI controller to do its thing then the onboard Promise controller to do its thing (as I have an IDE drive hooked up too).

From dead cold to desktop the whole process takes approximately 1:30. It never bothers me though...I walk into the MM labs...hit the big GO button...walk out...get a drink or use the bathroom...harrass the cat...when I come back I'm up and running.

But you are right...if you had to be up and running in 10 seconds...not going to happen.

All in all I'm glad that Tex hooked me up. It just has made the whole desktop experience a much more enjoyable and "smoother" place to be.

[t1rhino]

Let's say I had 6 brand new U320 15k 36gb hard drives.
For best performance, how should I set them up in the RAID controller?

[Tex]

Two or 4 channel? With 6 killer drives you can saturate two channels.

If you have only two channels put three on each and try a 64k stripe with raid-0. The optimum settings for your ontroller on how you set the caching and stuff can't really be guessed like this though as each controller is differant and they all tweak a little differant.

Tex

[Geeky1]

RAID 0... what stripe size is best is debateable; I just used the default 64k for mine. Maybe someone else can suggest an optimal stripe size.

Anyhow, RAID 0 is the best for maximum performance. The issue is, of course, if any one of those 6 drives dies, ALL of your data dies. RAID 0+1 (striping mirrored arrays) or 1+0 (mirroring striped arrays) (at least I think that's how those work...) should provide better data security without much speed loss...

[Tex]

The stripe for scsi raid with caching controllers is totaly differant then ide drives. There is nothing you can equate between the two.

Tex

[Geeky1]

Ya... didn't think about that...

[TheSmJ]

Here's the results for my WD800JB. Could anyone tell me weather or not the drive is performing as it should be?

[t1rhino]

Here is the benchmark of the latest toy.
There is next to no performance over my last benchmark. :scratch:

[Tex]

But rhino your maxing out a two channel card . 266 is about what mine will do maxed out and I bet yorus is close to that. With a 4 channel you can double that especialy caus eyour running pci-x right?

Tex

[Tex]

depends on how full the drive is. An empty one should hit 50,000 give or take on the rim. 3/4's the way into the drive you hit 35,000

Tex

[t1rhino]

All the drives are on one channel, and I don't really want to try and remove the server from the rack to change it.
I think it's PCI-X.
The drive only has a fresh install of win2k.

[Tex]

I didn't notivce they were u320. Sorry.

But the cache on that thing is really bluring reailty with atto.

You really need to run iometer to test a scsi rig like that. I was going to write and article about the differances ikn testing a scsi raid rig like ours and why you had to use IOmeter to really bench it in a server environment.

Cheers Buddy !

Tex

[t1rhino]

Should I try Sandra? Or another benchmark utility?

[Tex]

IO-Meter is the only true judge of scsi in a server environment. Is that what the machine is used for?

Tex

[t1rhino]

It will be in the future. ;)

[Tex]

IOmeter is what we used in the lab. You can set "number of workers" and attempt to set differant io patterns based on what they do... and it simulates differant load levels and stuff.

Tex