4x74gb Raptors or 4x250gb Maxtor's - Raid5 ?

Hi ppl, I'm recently upgrading my main fileserver from:

2x800mhz PIII
Supermicro 370DLE
1024mb ECC ram
9x9gb Atlas V U160 scsi drives on Adaptec 3200 raid controller (raid 5)
4x60gb IBM deathstars on Adaptec 2400a raid controller (raid 5)
Mandrake 8.2 I think (Yeah I know it's old)

to:

3ghz P4 northy
Asus P4P800 dlx (unused board was lying around so I figured cheap upgrade)
4x256mb OCZ3200
Possibly a Promise Fastrack S150 SX4 Raid 5 controller (I'm on a real tight budget here)
4xRaptor 74gb or 4x250gb Diamondmax 10 16mb cache drives
Mandrake 10.1

My question is: Will I notice the speed difference with the raptors over the Diamondmax 10's when accessing over my gigabit network?

It will mainly be fileserving for 4-8 Macs and a few PCs now and then. We mainly work with .PSD files ranging from 100mb to 1 or 2gb and have found the current (PIII) server cannot cope very well (high load 2-3 when disks are busy) mainly due to the Adaptec raid cards not working too well in Mandrake and giving me little over 40mb/s transfer speeds. I can get 50-60mb/s between any of my G5 macs on the same network with single SATA drives.

I think that the 200gb+ Raid5 volume from the raptors will be more than enough storage space for the next couple of years but given the same budget can get me to the magic 1tb of space I always wanted I presume the raptors are more suited to a server environment as well?

bt

EDIT: Changed the promise controller to the SX4 which I actually meant to put down and not the TX4, didn't know there were 2 types SW/HW versions.

Raptors are noticably faster than other drives, or so I hear. However, this is mostly in terms of seek time and not so much for sustained data rate (although RAID helps out there).

For a balance between storage and speed what about 2 of each? Have the most recently used files on the Raptors and the less used, archived, files on the Maxtors?

EDIT: SCSI drives would be better for a Server environment - being around as fast as raptors and with CRC checking and better data integrity measures.

Thanks for the quick reply David! Yeah I'm running raptors in both my PCs at home and know how fast they can be in raid0

I've ditched the scsi raid5 setup about 6 months ago due to a lot of corrupt data in some of our photoshop files. I don't know what was going on but I couldn't fix it so I bought the ATA Raid5 setup which sorted the problem out and was just as fast (or slow ) maybe bad drivers? I don't know Linux so it was left to one of the programmers at work to setup. Even in win2k server they performed terrible when tested.

I already got f***ed over twice by adaptec £600 and £400 for their 'server' raid cards and have used most of my budget already so SATA seems to be a cost effective solution to get me by for another 2 years.

David said:

For a balance between storage and speed what about 2 of each? Have the most recently used files on the Raptors and the less used, archived, files on the Maxtors?

Click to expand...

I really need the speed of raid5 over 4 drives and I only need one volume for current "Work in Progress" (100-150gb) only on this server as the old PIII server will then be used for the "Archive" (50-100gb) volume before archiving to DVD.

bt

I'd consider doing a raid 0+1 for what your looking at, you won't need an expensive controller then either.

Also IIRC that controller does a software RAID it's not a true XOR RAID 5. Real raid 5 controllers cost 200+

And in terms of performance, I would almost think two of those maxtors in RAID 1 would compete with those raptors in raid 5, and have more storage.

I'd checkout storagereview.com forums, they should have people who know much more about things like this.

Well, if it's only a server, who cares? A single raid card is not going to bog down a P4 3.0 (3 or 4 is a different story). Software raid doesn't matter at htat point.

ajrettke said:

I'd consider doing a raid 0+1 for what your looking at, you won't need an expensive controller then either.

Also IIRC that controller does a software RAID it's not a true XOR RAID 5. Real raid 5 controllers cost 200+

And in terms of performance, I would almost think two of those maxtors in RAID 1 would compete with those raptors in raid 5, and have more storage.

I'd checkout storagereview.com forums, they should have people who know much more about things like this.

Click to expand...

The SX4 uses a Promise XOR processor according to tomshardware but besides it will ONLY be used for fileserving and nothing else, the 3ghz P4 is totall overkill, I just need fast I/O so I doubt the CPU load will ever get that high.

So you say 0+1 is faster than a 4 drive raid 5? I always thought raid 5 was the fastest fault tolerant setup you could get. If 0+1 is faster then maybe going with 4 Diamondmax 10's is the way to go as this option with the raptors would not give me enough future capacity.

Also the P4P800 dlx has only 2 SATA ports and I really need a H/W raid that I can configure in the bios pre-os install. Getting cards that run smoothly under linux is an absolute must as I have had many a bad exp in that dept.

some more googling for me I think

bt

EDIT: Changed TX4 to SX4. Sorry I must have got confused over product nos, I meant the SX4 which has a XOR processor and not the TX4 which you rightly pointed out doesn't, neither does the TX4 do raid 5.

RocketRaid 1820 Benchmarks

Well, after a few months between posts (mainly due to me being rather busy) I now have the system up and running with some benchmarks.

After chating with some ppl at storagereview.com forums I finished up with a machine as follows: Ditched the P4P800 due to is gigabit lan sharing the PCI bus with the raid card.

Intel P4 2.8E Prescott S478
Intel 7201 Server Board, PCI-X, CSA gigabit lan
2x512mb Corsair XMS3200LL
Highpoint RocketRAID 1820A 8 Channel S-ATA Raid Card
Intel PRO/1000 MT Dual Port Server Adapter NIC
Antec 5U Rack mount case with Antec EPS 550w PSU
8x DiamondMax 10's 200gb S-ATA HDs

Benchmarks using Atto:

4 drives


6 drives


8 drives


8 drives Raid0


Just thought someone might want to see how different raid modes worked with 4, 6 and 8 drives. I was kinda expecting a little more but I guess it's not too shabby. It's gonna make a nice little samba server with 1.6TB of storage for very few pennies! I can't quite remember how much but I think it was around £1500 all in.

bt

Anyone know how to get these images to appear full size? They appeared full size when I put them on storagereview forums.

thats pretty cool to see.

i guess after 4 drives there isnt much increase in speed due to the speed of the pci buss. that sucks oh well. glad to see everythign worked out for ya

Wow 1.6Terabytes!!!!!

I shold start a poll for size of Harddrive array.

Yup, for most purposes those Diamondmax 10s are getting to be the pwnage of raptors. Raptors are only worth having anymore for single drive performance, and wierd situations where lots of really small widely scattered files are read. WD really ought to make a revised SATA native controller for them with RAID optimisations, to get them back in the game.

i just ordered a 300gb 16mb cache and i cant wait to match it up agains my raptor raid 0 see how well that works

Configured to 6x200gb Raid5 array using 2 drives as hot spares as this is how I will probably go with. Benchmark result from diskspeed

Sequential read
C:\>diskspd -h -d60 -o16 -b1M E:\test.tif
Test area size is 998 MB (using 16 overlapped I/O's)
WARNING: Amount of data processed exceeds test area size.
CPU 0 usage = 13.0%
CPU 1 usage = 2.5%
13291.00 MB in 60.00 seconds: 221.51 MB/sec , 221.51 IO/sec [avg. CPU 7.7%]

Sequential write
C:\>diskspd -h -d60 -o16 -w -b1M E:\test.tif
Test area size is 998 MB (using 16 overlapped I/O's)
WARNING: Amount of data processed exceeds test area size.
CPU 0 usage = 36.0%
CPU 1 usage = 1.1%
10011.00 MB in 60.01 seconds: 166.83 MB/sec , 166.83 IO/sec [avg. CPU 18.6%]

Random read
C:\>diskspd -h -d60 -o16 -r -b1M E:\test.tif
Test area size is 998 MB (using 16 overlapped I/O's)
CPU 0 usage = 4.9%
CPU 1 usage = 0.8%
6280.00 MB in 60.00 seconds: 104.66 MB/sec , 104.66 IO/sec [avg. CPU 2.9%]

Random write
C:\>diskspd -h -d60 -o16 -r -w -b1M E:\test.tif
Test area size is 998 MB (using 16 overlapped I/O's)
CPU 0 usage = 10.9%
CPU 1 usage = 0.7%
2761.00 MB in 60.00 seconds: 46.01 MB/sec , 46.01 IO/sec [avg. CPU 5.8%]

13.2GB in 60secs! That's nearly 1GB every 4 seconds Not bad for a fault tolerant raid5 setup, I'm pretty happy with it now

Mind you if I could get my backup server also with 3xgigabit nics a raid array that could write at >200MB/s it's only gonna take 75mins to back up 1TB! Man, that is a LOT of data to move!

Times have definately changed since the days I used to back up my first ever 20MB external HD with a box of floppies! lol

bt

if it is a file server you want space and those drives will DEFINETLY serve up your files fast enough

As said i think Raptors are best in a single drive or raid 0 configuration for programs and apps to run from, not for file storage.

The speed on that 8 drive RAID-0....... Kind of like the top fuel dragster of storage, fast as anything, lots of chances for something to crash and burn.

VERY helpful info for me...thanks for posting your findings

el said:

Wow 1.6Terabytes!!!!!

I shold start a poll for size of Harddrive array.

Click to expand...

Naah, not even 1.5TB when formatted with a Raid0 array Will it be enough for my needs?


Here are some pics of me fitting 8 drives plus a test winxp/Linux Mandrake10 HD in the same box, still room for a couple more drives yet though! All of the drives at the moment are cooled by 120mm and 80mm fans, if you've ever stacked 4 drives on top of each other you'll know it generates quite a bit of heat when being worked hard. Although obviously the DM10s feel a lot cooler and sound fairly quiet compared to the 10,000 rpm raptors I have in my PC at home.

Needless to say, even with the 3x80mm fans running @7v and 120mm @12v (it won't even move at 7v!) this thing is too loud for your desktop! Good job it's going in the air-conditioned server room. (Back to 12v all round in there)

The tidy angle!


The messy angle!


If I'd taken the pics with f22 and not f4.0 (not used to these digital SLR camera's yet!) you would see I've made stickers for all the cables and drives. It's definately worth the effort, later down the line when you get a S-ATA drive fail you dont need to guess what one to replace! Believe me, you don't want to disconnect the wrong one under a Raid5 array, you'll loose all your data if the spare hasn't re-generated in time! That was the advantage of scsi, you could send a command to blink the LED on each drive just to make sure before you disconnect the failed drive.

You don't want to remove the wrong drive when you've got a disaster recovery to do, if you get it wrong and lose the data the company will probably go out of business. I think I read somewhere around 80% of businessess go bankrupt after a total loss failure.

To simulate a HD failure, I disconnected the power to one of the drives in the array, I heard a few beeps from the controller, received an email reporting the fault during which time one of the 'hot spares' took over and started to re-generate all within 5-10 seconds of the power failure Re-generating the (1TB) raid5 array took just under 2 hours to complete. Whilst re-generating I also ran the Atto HD benchmark to simulate a heavy workload and still managed to achieve 130mb/s reads and around 60mb/s writes, more than acceptable in my view to keep our company working sufficiently during a raid5 re-generation/HD failure.

Overall I'm pretty happy with it, ok it would be nice to see it perform under a 133mhz PCI-X just to see if this 66mhz PCI-X slot is holding it back to around 220mb/s, after all, we'll always want something faster, that's why we're all here! I guess I'll never know, for now it will do it's job nicely

Stumpjumper5200 said:

The speed on that 8 drive RAID-0....... Kind of like the top fuel dragster of storage, fast as anything, lots of chances for something to crash and burn.

Click to expand...

To be honest, the raid0 array doesn't perform as good as I thought it would, the 6 drive raid5 array is nearly as fast as the 8 drive raid0 and also has the added bonus of fault tolerance + 2 hot spares! It just shows that for raw STR (sustaned transfer rate) it really does perform exceptionaly well especially compared to the (more expensive) competition.

bt

1.5 TB of pRon!

fatguy said:

1.5 TB of pRon!

Click to expand...

I wish! lol

bt

Wow,I am impressed that the raid5 suffers little speed loss compared to the raid0 4x array.I guess the raid0 8x is limited by the 4 channels while the raid5 seems to benifit from the 4 extra drives.I am curios as to how the acess times compare to other drive setups like a raptor and if the 4x,8x helps or hinders the acess times.

shadowdr said:

Wow,I am impressed that the raid5 suffers little speed loss compared to the raid0 4x array.I guess the raid0 8x is limited by the 4 channels while the raid5 seems to benifit from the 4 extra drives.I am curios as to how the acess times compare to other drive setups like a raptor and if the 4x,8x helps or hinders the acess times.

Click to expand...

It's a SATA connection, drives don't share connectors like PATA.

Off-topic: Bobtod - Nice! What kind of camera are you using? I just recently got a Digital Rebel, which is now trying to bankrupt me with temptations of lots of accessories and L-series lenses.