Re: [gentoo-server] RAID

[kashani]

Sean Cook wrote:

> My post did say "Back in the Day" and it was around 1999 that we did
> this however, then I was running unstable raid tools and they have come
> a long way, and I was running 2.2 kernel.

Hey I'm sure you saw what you saw, that's not a really an issue for me. 
However the link you posted was a study in "how not to do serious 
hardware testing."


> However, the only people I am aware of that say the performance is
> better on hardware raid are hardware raid manufacturers and you... You
> don't work for LSI do you?  Most of the linux software folks agree with
> me or take a mildly more conservative tone.

	Oddly this is the first time I've ever heard of software RAID being 
faster than real hardware raid... I've never seen any numbers that 
remotely point to such a case in my own testing or anyone that attempted 
to do real testing.

If I worked for a hardware vendor I'd hope to be getting paid more than 
I do at Internet startup #5. In any case I think you're misunderstanding 
the issues. Let's take a look at these links below.


> Redhat Enterprise Linux (circa 2005-2006)
> http://www.redhat.com/docs/manuals/enterprise/RHEL-4-Manual/pdf/rhel-isa-en.pdf
>
> Often the excess CPU power available for software RAID parity
> calculations greatly exceeds the processing power present on a RAID
> controller card. Therefore, some software RAID implementations
> actually have the capability for higher performance than hardware RAID
> implementations.

This is true, but does it actually mean you get more performance?

	I'd argue that any self respecting RAID card has more than enough CPU 
to handle saturating the slow in comparison drives. You can argue that 
they do not and that might be valid. If you're right then that RAID card 
does indeed suck ass.
	Assuming I'm right we can say that the CPU is not a limiting factor in 
either software or hardware raid. On the other hand I can say that since 
I have dedicated processing power for my RAID that I don't have to worry 
about spikes in load causing I/O issues as well as processing issues. 
Admittedly with today's processors you'd have to be taking a gigantic 
bite out of your CPU to affect your software RAID.


> Mysql:
>
> Hardware Versus Software
>
> Some operating systems can perform software RAID. Rather than buying a
> dedicated RAID controller, the operating system's kernel splits the I/O
> among multiple disks. Many users shy away from using these features
> because they've long been considered slow or buggy.
>
> In reality, software RAID is quite stable and performs rather well. The
> performance differences between hardware and software RAID tend not to
> be significant until they're under quite a bit of load. For smaller and
> medium-sized workloads, there's little discernible difference between
> them. Yes, the server's CPU must do a bit more work when using software
> RAID, but modern CPUs are so fast that the RAID operations consume a
> small fraction of the available CPU time. And, as we stressed earlier,
> the CPU is usually not the bottleneck in a database server anyway.

I understand this to mean that software raid can be just a fast as 
hardware raid when there are no limiting factors. This is also true.

Notice that neither of the above said software RAID was faster only that 
it could perform as well under the right conditions and save you money. 
I agree with that as well.

	However let's talk about a system that lives outside this perfect world 
that may have RAM, CPU, and I/O contention. A system where having a 
256MB cache on your RAID card turning expensive short reads and writes 
into nice long reads and writes by acting as an I/O buffer eliminates 
painful and expensive IO waits on your system. Things that you can 
actually measure in detail using iostats from the sysstats package while 
watching system load fall from 12 to .5.

In my experience having hardware raid cards has allowed me to get more 
per web, db, app, or mail server than software RAID resulting in more 
work for less money even with $ up charge for the raid card.

I'd be willing to use the spare Dell 2850 with five drives and a raid 
card to do some testing this weekend, but someone will need to offer me 
a free sushi dinner or two if I'm right. Any takers?

kashani
-- 
gentoo-server@g.o mailing list