Tough descision - RAID configuration (RAID5 or RAID0&RAID5)

[kokoon]

hello!

i'm setting up a system for a friend. we've been discussing this over and over and are having hard time deciding what to go with.

the machine will be used for two things:
- primarily graphic design (photoshop, flash, ...) and sometimes some video editing (premiere at most)
- secondary will be digital audio workstation - multitrack recording in cubase, running some softsynths and samplers.

we've already decided to have 2 separate windows installs, one for each. storage space will be shared (although futher partitioned)

now the hardware. the hardware is all fresh new.
- ASUS P5B Deluxe (that's Intel ICH8R matrix raid controller - supports RAID levels 0, 1 and 5)
- Core 2 Duo E6400 (works really nicely @2800 with stock fan)
- 3x SATA Western Digital Caviar WD2500KS (16 MB cache, 7200 RPM)
- 2x 1 GB Geil 800MHz

one of the main goals is to ensure high data safety (redundancy) but of course speed also. my initial idea was to go with this mobo and have just one RAID5 array. so i've done that, i set the block size to 128k and partitioned it to 2x50GB (the two system partitions), 3x120 (data partitions) and another partition for windows pagefile and various swap files from what's left (~20GB)
by the way - what are your opinions on pagefile location? how do you guys handle that?

i installed winXP and after some tweaking (enabling write-cache and such) everything seemed to run pretty smoothly. i can't really say how fast it was cause i don't know what it would be without the RAID and the install was fresh and wasn't bloated with tons of apps. disk benchmarks showed nice speeds for buffered operations and sequential reads, only writing was a bit lower compared to standard SATA setup (judging from the pc wizard benchmark)
the first question arises - how fast should a RAID5 over 3 disks be compared to a standard (non-RAID) setup? the way i see it: RAID5 is essentialy a striping array with overhead on writing (parity block needs to be calculated) but reading should be in the range of 2-striped RAID0 configs if not even better)
second question - CPU load: obviously the ICH8R doesn't calculate the parity bit, the main CPU does. should that bother me? correct me if i'm wrong but CPU only gets loaded with writing operations? what's the general consensus on the ICH8R RAID5 performance? good for general desktop use?

thinking about those 2 questions and browsing this forum i got another idea: this Matrix storage magic, i could set up a RAID0 slice (~100GB) and install the systems there and leave the rest in RAID5 - all sensitive data would be stored on the RAID5, still fast enough and backed up, while the system would run off a 3-striped RAID0. but what if one of the disks fail? both systems would disappear and i couldn't get to the "storage". so in that case i'd have to replace the disk, set the RAID0 back and reinstall the systems. at least one to get to the data on RAID5 slice.

and what about RAID5? if i had everything on RAID5 - how easy is it to recover from a crashed disk? the system should work with just 2 functional disks but slower, right? (one third calculated from parity blocks) so i could replace the disk, boot into the system and rebuild the RAID5... is that how it really works? anyone seen this in real life yet?

so - to summarize - what i'm asking you:
1. opinions on two separate installs
2. swapfile location - own partition or not?
3. RAID5 performance compared to no-array setup
4. ICH8R RAID5 implementation - CPU utilization: acceptable or not?
5. RAID5 reconstruction
6. final, the most important question, should i go with RAID0&RAID5 or stick with just RAID5.

answers on any of the questions would be really appreciated, i'm really inexperienced in this field and i'm doing this for myself... if it'd be for myself i'd try stuff and then switch to see but this way i have to make the descision...

please help!

 

[FeuerFrei]

1. opinions on two separate installs

Lost time configuring / switching / maintaining two installs

2. swapfile location - own partition or not?

I tried both, didn't made a real difference

3. RAID5 performance compared to no-array setup

Much better, and your data are safe(er)

4. ICH8R RAID5 implementation - CPU utilization: acceptable or not?

On a 6400? ROFL - (yes )

6. final, the most important question, should i go with RAID0&RAID5 or stick with just RAID5.

Just RAID5

 

[Randyman...]

Onboard RAID-5 or RAID-6 with a ICH-7 or ICH-8 will likely be PAINFULLY slow at Writes (slower than a single HD by itself). IIRC, the Write speeds were in the 20MB/s range on a 4 Drive RAID-5 on a ICH7R (recalling from memory). That is pretty slow. Reads will be insanely fast - but a DAW will need fast writes as well.

As you (and I) know, slow writes are a recipe for disaster on a high-track count DAW (unless you are not TRACKING on this machine). If you are simply IMPORTING projects for Mixing, then this is not a huge deal.

I am also a big DAW guy (I have 3 PC DAW's and a Roland 2480), and I have decided if I want a RAID-5 setup, that the only avenue is a GOOD Hardware XOR RAID card. I am eying the Areca 1210, 1220, and the 3-Ware 9650SE 4 and 8 port models. Price range from $300-$520 JUST for the controller.

That is where I'm at...

 

[kokoon]

damn. you're right about recording

i'm thinking getting another disk and setting up RAID10 with ICH8R... that shouldn't be any slower than RAID0 on two disks right?

 

[Casix]

damn. you're right about recording

i'm thinking getting another disk and setting up RAID10 with ICH8R... that shouldn't be any slower than RAID0 on two disks right?

It shouldn't be any slower yes, and it should be allot faster than the raid5 you now have on the ICH8R.

Basically the 2 best options are to get another disk and put it in raid 10, like you suggested, or buy a hardware raid controller and put the 3 disks in raid 5, but the last option is kinda expensive and i dont know if you want to spend that much money.
Just to give you an idea how much slower raid5 is on the ICH7R (the ICH7R is almost the same as the ICH8R) compared to a hardware raid controller see this: http://www.hothardware.com/viewarticle.aspx?page=10&articleid=869&cid=3
Also very interesting for you to read Randyman
So as you can see if you really want good performance your better of buying a hardware raid controller, although 4 disks in raid10 on the ICH8R should also give acceptable performance.

 

[greyharte]

hmmm ... im also thinking that i read somewhere a 128k stripe is too large as well, 16k seems to be the most optimal both for performance and reliability ... but then i could be wrong

 

[Casix]

hmmm ... im also thinking that i read somewhere a 128k stripe is too large as well, 16k seems to be the most optimal both for performance and reliability ... but then i could be wrong

It depends what applications you are using, you cant just say that one stripe size is the best.
If you do allot with large files your better off with a larger stripe size and with small files a smaller stripe size.

 

[kokoon]

cheers guys, thanks for all the knowledge.

what about the disks used in the array - how critical is it to have identical disks? it should also work with different disks right? what are the limitations? i mean if a disk fails after 2 years i probably won't be able to replace it with an *identical* one...

 

[eggi_1974]

I have the P5B dlx and use 4 seagates 7200.10, I use intel matrix raid and setup 5gb of each disk in raid0 and the rest in raid5. I have the machine on 24/7 and I use acronis true image to back up my OS every night.
I have run HDtach and attached the result.

 

[Casix]

I have the P5B dlx and use 4 seagates 7200.10, I use intel matrix raid and setup 5gb of each disk in raid0 and the rest in raid5. I have the machine on 24/7 and I use acronis true image to back up my OS every night.
I have run HDtach and attached the result.

Can you also run ATTO Disk Benchmark, because the HD Tach free version doesnt let you see write speeds, and thats what is so slow on the ICH8R with raid5.

 

[eggi_1974]

ATTO shows really poor write speed on the raid5 array(disk E), but in use it does not feel that slow.

 

[eggi_1974]

This happens when I turn off direct I/O in ATTO

 

[greyharte]

It depends what applications you are using, you cant just say that one stripe size is the best.
If you do allot with large files your better off with a larger stripe size and with small files a smaller stripe size.

thank you for that clarification

 

[eggi_1974]

I did a file transfer test, copied a folder with 6 video files with a total of 2.04GB

raid5 -> 15k scsi 27 sec

15k scsi -> raid5 126 sec

 

[kokoon]

This happens when I turn off direct I/O in ATTO

i don't get it. you get tons better speeds without direct I/O? am i reading this chart correctly? this is like megafast...

 

[Casix]

i don't get it. you get tons better speeds without direct I/O? am i reading this chart correctly? this is like megafast...

I also dont get it, but 1 thing is for sure that the benchmark with direct I/O off in ATTO doesnt seem right if you look at his file transfer test where he copied 2.04GB, the benchmark with direct I/O on looks more accurate, so its better to compare it to that.

cheers guys, thanks for all the knowledge.

what about the disks used in the array - how critical is it to have identical disks? it should also work with different disks right? what are the limitations? i mean if a disk fails after 2 years i probably won't be able to replace it with an *identical* one...

If you are gonna use raid10 you need identical disks, i'm not sure about raid5 though.

Wait i just found something about raid5:

Lets say you have 3 x 72gb hard drives and one of the 72Gb hard drives
dies

If you replace it with a 300Gb hard drive only 72Gb of it is useable -
the rest is wasted - but at least you are back working

In raid 5 you will perform at the speed of the slowest disk in the raid
set as it has to read the stripe off the disks o reconstruct the data so
one slow disk makes all the disks slow

Pretty much you can mix and match but the rule of thumb is it will work
at whatever is the lowet common denominator.

Hope this helps you out a bit

 

[kokoon]

why do i need identical disks for RAID 10? why is it different from RAID 5?

 

[hafa]

why do i need identical disks for RAID 10? why is it different from RAID 5?

You'll get better overall results using identical drives.

Regarding RAID 10, here's an Atto bench of 4 Seagate 7200.1s in RAID 10; 64K stripe on an Intel 632xESB Raid Controller (also matrix):



As you can see, it has phenomenal write times compared to RAID 5. Just as a curiosity, I also ran the same test on the system drive, as single new Raptor 150:



The read times on the Raptor really smoke the RAID 10 Array in the Atto benchmark, but HDspeed and HDtach tell a different story:




RAID 10



RAPTOR



RAID10



RAPTOR

I know that this is not really about Raptor VS RAID 10, but I thought you might find the results interesting, especially in regards to the different results from the different benches.

 

[kokoon]

no i'm really glad you posted these, i needed to know how fast RAID10 is. it's writes are really fasts but reads aren't THAT fast. i wonder why writes are faster.

still, what would happen if i put different disks in a RAID10 array? will it even work? how slower will it be and why? you guys have any links?

 

[greenmaji]

writes are faster because, there isn't any party data calculation being done, just copy data to one set stripe to the other.

Raid5,3 and 6 require party calculations (like the calculation used with ECC ram)