S.W RAID vs RAID accelerator (FastTrakR TX4310 ) @ RAID 1
 

We intend to add RAID capabilities on Lenovo M58 .System :Linux based on LFS kernel 2.6.25 , Intel Core 2 due ,2x500GB SATA 2 HDD.In our application we have many read/write operations on small files ~ 100KB ( mails ).We are looking for a benchmark on RAID 1 : S.W vs RAID accelerator based on FastTrakR TX4310 .

A few notes here:

Don't install your root partition as software raid. It will not work and if it does, you will have a hell of a time messing with it.

Promise cards + Linux = :(

I wouldn't trust Promise on linux.

Check out 3ware 8006-2LP. Cheap, easy, supported by EVERYONE, and it works. :)

That is for raid1. As you want more complicated configs / more arrays, you will need much more expensive cards. 3Ware 9650 is a great PCI-E card (several different models as far as ports numbers go).

But yea, I hope this answers your question. If you are thinking between software raid and a FastTrak, just go software and don't bother spending the $60 on a card.

------------------------------------
Michael Piekarski
Network Engineer
mpiekarski@hostmysite.com
www.hostmysite.com

I'm with mpiekarski on this one. Mark down another vote for 3Ware HARDWARE based RAID controllers..
If you don't want to spend the money on a true RAID controller, then go with software RAID


the Promise TX4310 is a fakeraid controller.. not worth wasting time or money on imho. Follow that link to see how one guy really feels about fakeraid.. heh

Good luck. As you can see in this thread so far, it is easy to find opinions and hard to find benchmarks.

I have benchmarked Promise fake raid 1 vs. no raid in Windows XP 32 in a large repeatable task that does about ten time as much reading as writing on a very large collection of small files. The task appeared to be 100% CPU bound in all cases and the fake raid was about 10% slower in each of several tests.

I don't really understand. Maybe it really was spending 10% of CPU time in the fake raid software (but even though I hate fake raid, I can't believe it is that bad). The extra bus activity for doubling the number of writes should have some impact on the CPU speed, but that shouldn't get anywhere near 10%.

In theory, a high ratio of read to write should make fake raid 1 faster than no raid, because it can reduce seek time by distributing reads between the two drives. From other testing, I think that doesn't really happen in Promise fake raid 1 in XP 32. I think the logic for distributing the reads is so flawed it is useless. But in the controlled tests described here, the overlap of CPU use and disk use was so complete that a moderate change in seek times would be invisible in the bottom line.

I have used software raid on Linux, and find it a lot easier and more flexible than fake raid. I expect Linux software raid also has better performance than Windows fake raid. But I never did a benchmark.

I haven't used hardware raid since many hardware generations ago, so I have no real data (as opposed to just opinion) on that. But my opinion is that it is not worth the cost, complexity, increased risk (raid controller itself is a significant point of failure), and increased recovery problems (compatibility issues during recovery). I would (and do) use software raid instead.

Interesting. I've found the opposite to be true..

Hardware RAID controller dies, remove it and put a replacement in. Boot the server. Array is back on line.. A decent controller should pull the config back in from the drives automatically, some others may force you to restore the config from a disk you made when you configured the array.

Drive in the array fails ? Unplug the drive while the system is running and plug in the replacement drive. (Hot SWAP). Array rebuilds automatically.. what could be easier ?

Want to expand the array ? Plug in new drives while system is running, go into array management and add the drives into your defined array and tell it to grow.. Array rebuilds accross all drives assigned to the group..

I don't see how that is complex at all, maybe things have changed considerably for the better since you last used it :)

I'm currently considering moving from (linux) software raid to hardware raid. But I am concerned about what people say about dying raid controllers. A lot articles about raid mention the possibility...

Are hw raid controllers that fragile?

@farslayer, sounds like you have experience with dead raid controllers? Do you keep spare raid controllers for when dies? Do you have to replace the raid controller with the exact same type/model?

Thanks for any input!

Robbert

Replacing a Hardware RAID controller with one of the same make and model is convenient, although several manufacturers have eliminated the necessity for this. say replacing a 4 port Hardware RAID controller with an 8 port from the same manufacturer, or even replacing an older model card with a newer model, and it should detect the array just fine and operate normally, even with the disks in different positions. .

A most excellent question to ask the manufacturer before purchase. Tech support is not just for when problems occur, it's a great resource before you purchase a product, just for this sort of question.

Personally I've never run into the issue of replacing a hardware RAID controller that failed and losing the array.. Provided the data on the disks was not corrupted when the controller failed. I have had a multiple disk failure on an array, and well if you loose too many disks it really doesn't matter if the array is software or hardware based, it's time to restore from your backups.. ..

In the above mentioned incident, a drive failed in a RAID 5 array. the array had 6 Drives + 1 hot-spare, so 7 drives total in the enclosure. When the disk failed the system automatically inserted the hot-spare into the array, and started the rebuild, at which point 2 more drives failed during the rebuild process.. it was pretty much unrecoverable at that point. What a fun Monday morning that was..