Sign in with
Sign up | Sign in
Your question

slow write speeds on LSI RAID controller card

Last response: in Storage
Share
March 15, 2012 7:24:52 PM

running Win7 64bit, 12GB RAM, Core i7 930 stock, 64GB SSD for OS. I have an LSI 9240 8 port RAID card and a 9260 8 port RAID card with 512MB on board RAM. 4 x 2TB 5900 RPM SATA 2 Seagate hard drives and 2 x 2TB 7200 RPM SATA 3 Seagate hard drives, all in a single RAID 5 on the controller card. I've tried both controller cards and I'm getting really poor speeds, especially write speeds, dropping down to 14MB/second writing large video files over. I've ran tests on the hard drives and the RAID consistancy and I can't find any problems. Updated drivers, updated firmware. A friend of mine said it is because of the different speed hard drives, but googling it, all the forums I've seen say the same thing, that the RAID should perform as fast as the slowest hard drives, so they should be about the same as 6 x 2TB SATA 2 5900 RPM hard drives. Anyone have an opinion or any help?
CrystalMark speeds ~225MB/sec read, ~30MB/sec write.
I've also tried installing it into my gamer system. Core i7 2600K OC, 16GB RAM. Same speeds. Tried both sets of cables on both RAID cards. I read online somewhere that some motherboards don't like anything but graphics cards being installed in the PCI-e 16x slots. I've talked to MSI about this and they can't confirm or deny this.
a b G Storage
March 16, 2012 11:01:53 AM

It's because you're running RAID 5, RAID 5 is known for it's bad write speeds because fo the partiy calculations. Take a look at the link below, it looks like ~30MBps might be all you can expect from that card in RAID 5 for write performance.

https://docs.google.com/viewer?a=v&q=cache:KaGmvg2ihe8J...
March 16, 2012 12:34:50 PM

Thanks for getting back to me. I was told that the new firmware and drivers fixed that issue. I also have a 9260-8i, which is a hard drive RAID with on board RAM that is getting the exact same speeds. Even my read speeds are really poor.
Related resources
a b G Storage
March 16, 2012 10:31:15 PM

You'll never get close to 1/2 that speed, those tests were done using 15k SAS drives.. Your drives are SATA and rotating at 1/3 of the speed as the ones used in the test and you're using 1/2 as many disks. Just donig the quick math you should be at no more than 1/6th the speed of those tests. Your results are a little slow, but not an order of magnitude off from where they should be. I run 4x5900 RPM drives in RAID5 and I see approcimately the same speed you do, 30-35MBps and thats on a factory configured NAS device.
a b G Storage
March 16, 2012 10:32:23 PM

BTW I misread your original post, you have 6 drives in RAID 5, but keep in mind you get no benefit from the 7200 rpm drives, the array will operate as iff the disks are all 5900RPM
March 16, 2012 10:56:12 PM

yea, I knew that I wouldn't get a speed boost off of the 2 7200RPM drives. I picked them up for the same price as the 5900RPM drives so I figured I'd grab them and swap them out as I do other builds along the way.
Other people seem to think that I should be getting much better speeds than this, including LSI themselves. They don't know why I'm getting such poor speeds. They just keep putting the blame on these being desktop hard drives instead of enterprise hard drives saying that enterprise hard drives have a lot less time out than the desktop hard drives. That I understand, about time out errors. But I am rarely getting them. But they are saying even at the same speed enterprise drives I should be getting a lot faster write speeds.
I also have a HighPoint Rocket RAID 4 port card which the 4 5900RPM drives were connected to. The HighPoint is a pure software RAID card and I was getting close to 200MB/sec write speed on it.
a b G Storage
March 16, 2012 11:38:54 PM

Do you have writeback cache on your controller and is it enabled?
March 16, 2012 11:45:02 PM

not sure what that is. I know it has 512MB on board on the 9260. I'd assume it would be enabled. It shows in the cards BIOS. The 9240 doesn't have on board cache and is a software RAID. The 9240 and the 9260 is getting the same speed. The RocketRAID didn't have on board cache and was a software RAID card and was a lot faster.
Thank you so much for your help with ideas. I'm pretty technical and used to work at IBM, but I always worked on desktop and laptops. The RocketRAID was my first real RAID 5 set up. It worked great, I just ran out of room and thought the LSI would be better. I should have stuck with HighPoint
a b G Storage
March 17, 2012 8:09:26 AM

http://www.lsi.com/downloads/Public/MegaRAID%20Common%2...

http://www.lsi.com/downloads/Public/MegaRAID%20Common%2...

I would try upgrading to the latest firmware and drives to start. The firmware was updated on Jan30th and the driver is from March4th so they're pretty recent. Next download the management utility, you should be able to check to see if write cache is enabled. If you only have read cache or the majority of memory dedicated to read and not write, that could explain the results you're getting. Cache will make a huge difference in performance.

http://www.lsi.com/downloads/Public/MegaRAID%20Common%2...
March 17, 2012 8:42:42 AM

I don't know about your controller but the best write speeds on raid5/6 are obtained when the write cache is enabled, a lot of cards will not let you enable this without having battery backed up cache installed

My 8 disk raid5 array on one of my controllers have terrible speeds with no BBWC , with the BBWC installed it handles large file writes from 70 to 100mb/s
a b G Storage
March 17, 2012 5:38:32 PM

Yea I think he is not using write cache... write performance will be awful without it.
a b G Storage
March 17, 2012 6:16:25 PM

Mac_angel, is there a battery on your RAID controller? If not is there a connector for a battery?
March 17, 2012 6:27:46 PM

no battery on the RAID card. Yes, there is a jack for it for the 9260. I'm running the latest drivers, firmware, and software. I'm pretty sure the on board RAM is working, though I'll have to mess around and double check. The RAID card also supports Cachecade, being able to also use an SSD for a cache file. I don't mind picking up an inexpensive one, but I want to make sure I'm actually going to get enough speed out of it. This is only for home use, ripping my DVDs and BlueRays to access around the apartment. Once all the movies are on there then I only really need read speeds. But it's taking a long time to put the movies on there.

again, thanks for all the help in trying to trouble shoot this. I'm a tech geek myself, but desktops and laptops. I was a technician for IBM, actually. The HighPoint RAID card was my first dip into a dedicated RAID card. I'm stumped. Does anyone think that it could be the mix of hard drives, 4 x 5900RPM and 2 7200RPM?
March 17, 2012 9:07:40 PM

okay, I was going to take the day off from this puzzle to relax, but I thought I'd mess around with it for a few minutes. I've been 'forced' into retirement because I have EDS. Think of a bad case of arthritis, so for the past while, messing around with the computers and moving around a lot has killed my knees lately.
Anyway, I just wanted to say that I uninstalled and reinstalled everything. Drivers, Firmware, and software. I'm getting ~235MB/sec read speed and ~175MB/sec. Not perfect, but much better, thank you. Seems like the cache wasn't set up properly. There is advanced software for this as well, but it's not activated. I guess I have to call in to see about that.
a b G Storage
March 18, 2012 5:59:48 AM

Be careful if you have no battery backup and you're using the write cache. I HIGHLY recommend putting the battery in place, if you lose power during a write you can corrupt the whole volume. I'd glad things are working better for you.
March 18, 2012 1:09:25 PM

the computer itself is hooked up to a battery back up, but thanks for your concern and all your help. :) 
I'm going to swap out the 5900RPM drives over the next while with 7200RPMs. I'll probably get a good speed boost from that.
Odd thing that happened through my trouble shooting and trying different things, I was actually getting a read speed of ~335MB/sec. I'm not sure what I did that I lost that speed. If I got that speed up then I'd probably get the write speed up as well.
a b G Storage
March 21, 2012 8:07:28 PM

The speed you get from LSI 9240 is slow because it's NOT a hardware RAID card.
Look at the spec... there is NO I/O processor included.

This mean, the host CPU is doing ALL the calculation!
You may said, there is i7 or dual core, quad core CPU, it should be able to handle that...
But you don't really that each drive is sending 1000's of IRQ per sec direct to CPU, which puts a high demand on on CPU clock cycle. On top of that Window won't dedication much of CPU time to assist the RAID engine.

Not even mention: drivers, BIOS, conflict with other card...

As the rule of thumb when come to RAID5 - use HARDWARE RAID, dont cheat yourself with hardware assist RAID such a card

If you need about 200MB ~ 265MB/sec transfer raid then use SPM394 or SPM393

If you need about 450MB or more the use PCIe hardware raid like: ARC-1223-8i
a b G Storage
March 21, 2012 8:14:14 PM

tokencode said:
It's because you're running RAID 5, RAID 5 is known for it's bad write speeds because fo the partiy calculations...


That is incorrect!

It bases on the card raid engine that you used.. some are good some are just plain BAD.

But generalize "RAID 5 is known for it's bad write speeds" is not true. We have RAID5 arrays that read/write over 1100MB/s
a b G Storage
March 22, 2012 1:37:47 AM

Actually firewire, I don't care what RAID card you're using, that statement IS true. If you use RAID10 on the same RAID card it will be faster. Yes you can have RAID5 arrays that perform well, but if you used the same number of spindles, you well get better performance from RAID10. RAID with parity calcs (5 or 6) has worse write performance than RAID 0 or 10 with the same number of spindles all other things being equal. Go run a RAID0 on your card and tell me you don't get better write performance..... of course if you are using different hardware your results will be different, but if you isolate just the RAID level, RAID5 is known for it's bad write performance, don't be so quick to say someone is incorrect.
a b G Storage
March 22, 2012 1:38:50 AM

Please remember spindles is the number active drives used for striping, IE a 10 disk RAID 10 has 5 active spindles.
a b G Storage
March 22, 2012 3:55:32 PM

"IE a 10 disk RAID 10 has 5 active spindles."

Where 10 disk RAID5 has 10 active spindles :-)

Like i said, it depends on what RAID card you are using, how efficient the raid engine and how fast the I/O CPU
a b G Storage
March 23, 2012 3:13:48 AM

well 9 but I get your point, although a 5 spindle (10 drive) RAID 10 will be faster than a 9 spindle 10 drive RAID5 when it comes to writes, that is why RAID 10 is almost always recommended for all databases or other processes that require fast random read/write
a b G Storage
March 23, 2012 7:03:41 PM

Again 10x spindle in RAID5 - Not 9. LOL.
Each drive has 1/N portion dedicates as parity, where N is number of drives
That is why you loose ONE drive in RAID5. Example:
5x drives, each drive has (1/5) 20% capacity taken out
10x drives, each drive has (1/10) 10% capacity taken out
16x drives, each drive has (1/16) 6.25% capacity taken out

So when Read/Write, they all run.

There is NO dedicate parity drive in RAID5. Parity data spreads across among the drives
Think about this, what happen to the raid if the dedicate drives failed (it could happen), then there is NO parity data to rebuilt the volume :-)

Matter of fact for Database Server, which has read more write - it will benefit from RAID5 for fast read compare to RAID10

There is lot more on sector, buffer and other stuff setting in raid setting... But if you are correctly using it RAID5 will kick but RAID10 performance.

It matters of how you use/set it
a b G Storage
March 23, 2012 7:35:04 PM

That's not correct because not all of the data you are writing is actual data, N-1 is parity data therefore you only see the performance of 9/10 spindles because there is 10 additonal data to write due to parity.

Turst me, go try to run a highly utilized SQL server on RAID 5 and let me know how that works out for you....

Always place log files on RAID 1+0 (or RAID 1) disks. This provides:

better protection from hardware failure, and


better write performance.

Note: In general RAID 1+0 will provide better throughput for write-intensive applications. The amount of performance gained will vary based on the HW vendor’s RAID implementations. Most common alternative to RAID 1+0 is RAID 5. Generally, RAID 1+0 provides better write performance than any other RAID level providing data protection, including RAID 5.


http://technet.microsoft.com/en-us/library/cc966534.asp...

a b G Storage
March 23, 2012 7:46:01 PM

The above article reference is for log volumes, but the same applys for data volumes of a given size as well. If your database is primarily reads then yes you might get better performance with RAID 5, but if it is evenly read/write or primairly write, RAID 10 is definitely the way to go, here's a better explanation than I could give....



According to sql-server-performance.com, the best RAID configuration is dependant upon the number of I/O Reads versus I/O Writes. I have quoted the following from their website to help anyone else that may have the same question that I did.



When performance tuning a SQL Server, it is often handy to know if the disk I/O of your servers (and the databases on it) are mostly reads or mostly writes. This information can be used to calculate the ratio of writes to reads of your server, and this ratio can affect how you might want to tune your SQL Server. For example, if you find that your server is heavy on the writes, then you will want to avoid RAID 5 if you can, and use RAID 10 instead. This is because RAID 5 is much less efficient that RAID 10 at writes. But if your server has a much greater number of reads than writes, then perhaps a RAID 5 system is more than adequate.

One of the quickest ways to find out the ratio of reads to writes on your SQL Servers is to run Task Manager and look at the sqlservr.exe process (this is the mssqlserver service) and view the total number of I/O Read Bytes and I/O Write Bytes. If you don't see this in Task Manager, go to View|Select Column, and add these two columns to Task Manager.

The results you see tell you how many bytes of data have been written and read from the SQL Server service since it was last restarted. Because of this, you don't want to read this figure immediately after starting the SQL Server service, but after several days of typical use.

In one particular case I looked at, the SQL Server had 415,006,801,908 I/O bytes read and 204,669,746,458 bytes written. This server had about one write for every two reads. In this case, RAID 5 is probably a good compromise in performance, assuming that RAID 10 is not available from a budget perspective. But if the reverse were true, and there were two writes for every one read, then RAID 10 would be needed for best overall performance of SQL Server.
a b G Storage
March 23, 2012 7:46:49 PM

BTW, I'm glad we fixed the OPs probablem because we totally hijacked this thread.
a b G Storage
March 23, 2012 7:51:32 PM

:-) yes Indeed
March 25, 2012 10:50:35 PM

lol, yea, no kidding.
suggesting new RAID cards really doesn't help much when I said what I have already. And a software RAID isn't so much a problem with the kind of rig I have. Thing was, it wasn't using ANY resources at all. I still think I'm getting shitty speeds considering. It is a hardware RAID card with 512MB on board RAM, and it should be a fast enough rig, but 250MB/sec read and 175MB/sec write, yes, it's better write then what I had before, but no where near where it should be. Part I'm confused on is I was getting 335MB/sec read at the beginning, and it's now down to 250MB/sec. I don't know what happened to loose that speed. I've tried playing around with things in the BIOS and the settings, but no luck.
!