Software RAID under Win2k

[RichL]

Anyone had the chance to run a software RAID 0 setup versus a hardware one under Windows 2k?

A freind of mine has semi-promised me FOUR 9.1gb fast/wide ultra scsi 10,000rpm disks and a suitable Adaptec non-raid controller, and the thought of 36gb on a four disk Raid of high speed Scsi is too tempting to resist.
If only I could justify £500 for a scsi Raid card!

[WyWyWyWy]

You can only do Raid0 or Raid1 with the Windows inbuilt feature

[DGhost]

what i've heard is that if you are running a system with one (or more) high end processors, windows software raid can turn out to be faster than true hardware raid. granted, that was tested using IDE drives so i'm not sure how SCSI would compare... but... oh well...

[RichL]

I did see a bit on Toms Hardware that said on modern CPUs the overhead for software raid was often about the same as the overhead for running a hardware raid controller.
But this was on Toms Hardware..

[Slougi]

Tom'z Hardware Tribe?

[DGhost]

storage review acctually... link is here ... its just a bit old... like... 4 years old...

[RichL]

That reminds me - for anyone in the UK, you can get Promise IDE RAID controllers (TX2 and TX4) from Maplin Electronics

Its the only place I've ever seen them for sale, so I thought I'd mention it.

[luni]

Well all the popular low end IDE raid controllers like Promise or High Point are really using software emulation anyway, they only really provide a mean to attach those additional drives and driver instructions. All the hard work is actually done by your CPU...

I have never tested software raid in practical application, but on a decent CPU it should be same speed as low end IDE raid controllers and faster than vast majority of dedicated hardware raid controlers (at least in 0 and 1 modes).

Personally I think that even RAID 5 and others should be no problem for modern CPUs, if only they would implement the possibility in software. But I guess that would interfere with controller bussiness too much

[Wombat]

Got anything to back that up luni? I'm pretty sure this isn't just emulation. Doc!?!?!

[DGhost]

its true. this has been demonstrated time and time again with the driver hacks and bios hacks for the promise controllers and even some of the high point controllers.

the low end controllers are just IDE controllers with a custom bios and custom drivers that enable raid abilities by using the host processor.

its one of the reasons they don't support RAID5 (which does take a lot of processing time) and why they use more CPU time than dedicated hardware raid controllers... and why they also use more CPU time when more drives are added... and also why they are so cheap...

[Dilitante1]

my RAID5 (3x4gig) Win2K server runs nice

[UtwigMU]

You can implement software RAID5 using NT/2k Server, I don't know about 'nix.

I did a calculation and more expensive RAID5 controller adds to the cost of 4-drive RAID5 array in such extent that 4-drive RAID10 costs same per GB - IDE RAID. (Promise TX4 vs SX6000)

I also saw posts on software and hardware RAID arrays being screwed due to different drivers or misscheduling.

RichL - If going RAID IMO do it properly get a controller

Luni - prvi Slovenec, ki sem ga srecal na fourumih. Od kod si stari?

[rugger]

Software raid works fine. I have used software raid on production systems (under linux of course, both raid 1 and raid 5), with NO ill performance effects.

Don't base your decision to use hardware/software raid on the performance aspects.

Base it on the following:
* Do you want reasonable transperancy of the array across operating systems? If so, use hardware raid.
* Do you want advanced RAID features like spare disks, hot-swap and audible alert? If so, use hardware raid.
* Do you want to raid your boot device? If so, use hardware raid. (Win2000 won't boot from raided partitions)

[Dr Mordrid]

Won't use software RAID in Win2K for these reasons:

1. higher resource utilization, which is bad news for video editing...especially using realtime solutions that use the CPU to do some of the effects (most newer devices will do it this way)

2. limited redundency options in that you dont have RAID5, which forces you to use mirrored arrays.

Promises FT SX4000 fixes all three problems and at <$165 for a full hardware RAID5 with hardware XOR (the main reason for slow RAID5 in many previous cards is the use of software XOR).

Love mine.

Dr. Mordrid

[luni]

Wombat - DGhost summed it up pretty well. I also did some raid controller testing a while ago and found out that dedicated RAID controllers offer less performance than cheap ones, of course in standard 0 and 1 modes (I couldnt test others, since those two (and the comical 0+1) are only ones that cheapos offer). I was wondering about that, but when you closely examine the hardware controller, you realize that all bit distribution work is actually done by a really really really slow little CPU, in most occasions an Intel i960 chip. Now, it is true that that chip is dedicated only to array calculations, but modern CPUs are so much faster, that they can handle those calculations with little utilization, ie. much faster.

UtwigMU - Maribor . Sem ze prej studiral ce si iz Slovenije, ampak nisem hotel nic sprasevat .