Which one is better for my CentOS? Hardware RAID using Dell PERC 6 or Software RAID?

Question

I would like to setup RAID 0 for my Dell 2900 server. Which one is better, using hardware RAID (using installed Dell PERC 6 controller) or software RAID? The RAID 0 is for database.

Solution 1

Hardware RAID is the only way to go, if your pockets allow.

Solution 2

See my other answer regarding raid to get the concept.

Dell PERC 6 although a real hardware raid is a "cheap" one. You should avoid that one and use software raid instead (if you can). If you end up using it as a hardware raid no worries, it is not that bad. Just do not expect too much from it (e.g. bad performance). There are also binary utilities that can be used to manage the controller from linux. They all suck compared to good hw raid or software raid.

My comment is based on my experince with a 1 or 2 generation older card. It could have been improved since, but I do not think so.

Solution 3

It depends on what you know and what you buy. Back in the day, I managed FC arrays on Sun servers with Veritas Volume Manager and software RAID. Things worked great and performance was excellent. I had a similar experience with AIX 4 servers a few years later.

Hardware RAID should be faster; but in reality the trashy RAID controllers that you often find negate that, and you'd be better off with a maintainable software RAID and lvm. Mr. Atwood's experience with the IBM ServerRAID 8k isn't an isolated incident. We had an incident where a defective RAID controller firmware forced us to visit over 400 remote locations to perform a manual upgrade.

If you don't know whether your RAID controller is junk or not, test, test, test.

Solution 4

This Stackoverflow posting talks about RAID layout for database servers. You should not use RAID-0 as it has no redundancy. RAID-0 is interesting for performance at the expense of everything else; you might use it on a video editing system (for example) but it is not appropriate for databases.

Linux has much better support for software RAID than Windows does. I would almost never recommend software RAID on a windows server. Linux is a different proposition. The software RAID on Linux is quite good and is certainly worth considering. On Linux, a fast CPU might result in better RAID performance than a slower RAID controller, but most SATA or SAS RAID controllers have fairly beefy RISC cores with hardware XOR units. Some even have multi-core CPUs clocked at more than 1GHz.

Most Dell PERCs are rebadged LSI or Adaptec controllers. One point to note is that they can take a battery backed cache, which is a feature that is not supported on software RAID. This gives you extra resilience to certain failure modes.

My suggestion would be to run some benchmarks with bonnie++ or something similar on both H/W and S/W RAID. If either configuration is noticeably faster go with that. However, if the server is currently in use and not causing any problems you may want to stick with your current configuration. The configuration is not likely to make enough of a difference to warrant rebuilding a server that is already in production.

Solution 5

Running a software RAID seems kind of like asking a airline pilot to fly the plane and serve drinks and peanuts to all the passengers at the same time :-)

In my opinion it's always better to use hardware that's dedicated to a specific task, thus I would choose the PERC RAID controller over the software RAID, regardless of the OS you are running.

View more solutions

Comments

[BobbyShaftoe]

Despite what I hear Atwood saying on his podcasts; a proper hardware based RAID is going to outperform software solutions. Although, I wouldn't necessarily consider RAID 0 to be the best choice for a database server.

[squillman]

Agree with @BobbyShaftoe - RAID 0 is not the best thing for db performance. Do you need that level of redundancy?

[Element]

@Bobby: In defense of Jeff I don't think he ever said software raid would offer better performance, he theorized it would be more reliable because it avoided the poorly written firmware on raid controllers.

[Zoredache]

@squillman, what redundancy. RAID0 is striping. If any drive fails you loose everything.

[Oskar Duveborn]

You do mean a mirror, RAID 1, yes?

[Rob Bergin]

Maybe a typo - RAID 10 is best for DBs (RAID 5 is pretty good too).

[Rob Bergin]

Or a... RAID 0+1: striped sets in a mirrored set (minimum four disks; even number of disks) provides fault tolerance and improved performance but increases complexity. The key difference from RAID 1+0 is that RAID 0+1 creates a second striped set to mirror a primary striped set. The array continues to operate with one or more drives failed in the same mirror set, but if drives fail on both sides of the mirror the data on the RAID system is lost.

[Rob Bergin]

Or a.. RAID 1+0: mirrored sets in a striped set (minimum four disks; even number of disks) provides fault tolerance and improved performance but increases complexity. The key difference from RAID 0+1 is that RAID 1+0 creates a striped set from a series of mirrored drives. In a failed disk situation, RAID 1+0 performs better because all the remaining disks continue to be used. The array can sustain multiple drive losses so long as no mirror loses all its drives.

[squillman]

Aye. Hardware RAID = good.

[Oskar Duveborn]

What parts of Windows Server software raid is not up to par with Linux?

[duffbeer703]

Generally speaking with Windows Server sw raid, people are pointing and clicking to set things up, and that ease of use makes it likely that someone will screw it up. On the Unix platforms, there's usually a good & easy to use command line that can be scripted so that low-level techs can do stuff.

[ConcernedOfTunbridgeWells]

S/W RAID on Windows suppors a limited range of layouts (no RAID-10 or RAID-50 support as far as I can tell) and is dreadfully slow on RAID-5.

[demonkoryu]

Don't be so quick to dismiss CPU requirements. It's true that some boxes have spare CPU cycles, but there are also many cases where a box is running a CPU bound application and every cycle counts.

[Dan Carley]

Upvote for this. The opinion may upset some people, but a half-cut HW RAID setup is indeed by no means better than some proven SW RAID. Cards with cheap or ageing chipsets can be just as much burden on system resources whilst altogether more unreliable.

[Decebal]

yep, but fakeraid (ie BIOS based) is very bad. software raid is quicker and more reliable. HW > SW > Fake.

[Decebal]

if every cycle counts, those boxes are not writing to disk all the time. My linux web server uses less than 1% CPU for its software RAID1.

[Atulmaharaj]

I think you're mistaken about the recovery of software raid. Since it's not bound to specific hardware and firmware, it's usually much easier to recover on another machine than hardware raid. My experiences are limited to linux software raid, which works very well and can even be recovered from a random linux live-cd. It seems you're throwing software raid and 'fakeraid' onto the same heap. Yes, they're related but the advantage real software raid has over fakeraid is that no specific (crappy) hardware is required, hence less points of failure.

[Coops]

How to tell if fake, is there always going to be a noticeable board for hardware RAID? Additionally dont most motherboards have 'fake' RAID these days, in which case would you be better off with software RAID than that?

[Jürgen Strobel]

I couldn't agree less. I just had a real nightmare using a Perc controller a couple of years old with new SSDs. It turns out software RAID 10 over hardware RAID 0 for every single disk (JBOD is not supported!) is double as fast as hardware RAID 10. CPU load is below noise level.