Unsolved
This post is more than 5 years old
19 Posts
0
48072
Slow disk performance with SAS 5/ir controller
A customer got a two year old Poweredge SC440 with a SAS5/IR PCI Express card. The Windows 2003 server is used for the Microsoft Retail Management System RMS) (Head quarter) for a small retail chain (< 20 stores). While the server initially performed fine, for the past 12 months or so the server appears to be hanging frequently when a store or two connect to the server.
The SAS 5/ir is configured as RAID 1 with two WD2500YS GB disks. Due to the 'nature' of the SAS 5/ir Disk Caching is disabled (no BBU), write-ahead is also disabled. It seems every known performance increasing technology has been disabled for this controller. The controller simply 'can do' RAID 0, 1 ...
Since there is no battery power cache on the controller, MS SQL server recommends to do direct write through only ... no caching again.
Further analysis revealed that the MS SQL server is performing slow and when I simply did a copy of the 5GB SQL DB .mdf file, the server seem to be not responding to my request. Explorer.exe was not responding and other application started to hang too.
Then I used robocopy (part of the MS server resource kit) to copy the file on the command line level. The simple copy of the 5GB file from one directory to another directory on the same RAID 1 virtual disk to 57 minutes. I checked this on other server and PC and it was always faster than 57 minutes.
I then run the Passmark test for the disk only and found that the 'Disk Sequentail Write' was about 30 to 50 time slower than on my S390 workstation (1.63), which lowered the overall Passmark result for the disk to 134.1.
The SQL server seems to run into problems as soon as more than one store is connected to the RMS head quarter server. A trace revealed that read/write operations of the sqlservr.exe caused most of the delays, which made me believe that the disk access speed is the bottleneck.
Anyhow, after finding another post (http://en.community.dell.com/support-forums/servers/f/906/p/18410505/18533491.aspx) I followed the instructions on how to enable Disk Write Caching (on the disk, not within the controller) and the results were stunning.
Instead of 57 minutes for the 5GB copy, it only took 5 minutes and 15 seconds.
The Passmark test changed dramatically as well ... 'Disk Sequentail Write' went from 1.63 to 45.9 and the overall result moved from 134.1 to 310.1.
While I understand the dangers of the enabled write cache, I cannot understand that anybody would use a controller that slow.
This controller is simply unusable for any normal MS SQL server application!
I now need to install a 'real' RAID 1 controller with BBU.
Anybody out there who could recommend a hardware RAID 1 controller for the SC440?
Would the PERC 6/i work in the SC440?
brshoemak
2 Posts
0
July 24th, 2010 09:00
A PERC5 will work in a Dell SC440, but I cannot guarantee a PERC6 will.
You say that it performed fine for 12 months, then if nothing changed on the system, why would the storage controller be at fault at this point vs. something relating to software. If this machine was purchased for the explicit purpose of being used as a SQL server then at the very least some sort of baseline performance tests should have been run initially, without a baseline saying "it ran fine" doesn't mean much if you can't quantify it. The later results you post regarding the system have no point of reference as their are not initial numbers to compare it to.
Also, a SAS 5/6 RAID controller was never designed to handle a decent SQL load. I have a SAS6ir in a SC440 and wouldn't dream of running SQL on it, but I looked at the performance metrics of the card before I purchased it.
If you are looking at anything relating to SQL you should have a PERC controller at the very least with a higher RAID level that RAID1 (like RAID10), but to blame the storage controller for decreasing performance doesn't make sense to me unless your load and changed significantly and that load was never accounted for during the initial procurement process.
Also, a SAS5/6ir is a 'REAL' RAID controller, it just doesn't have the performance that should have been looking at in the first place.
To answer the question at hand, the PERC6 SHOULD work the x8 PCIE slot of a Dell SC440 it's not guarantee and Dell will never tell you if it will since it was not an original system option. I went through the same ordeal with the SAS6ir that was not an original equipment option.
Ricky42
19 Posts
0
July 27th, 2010 00:00
Thanks for the feedback!
The SC440 was purchased by the company which installed the MS RMS system and I am pretty sure they did not do any great tests before.
Initially a smaller amount of stores were connected to the server and the amount of items in the database was lower too.
After the database grow to a size of probably 3 or 4GB the performance issues started (12 months ago).
After tracing down where the bottleneck was, I came to the conclusion that the disk controller cannot handle the SQL server transactions.
It also looks to me that whenever the database has to be accessed close to the end of the database file (offset around 5GB), the server, explorer etc. are hanging.
Unfortunately the server seems to be good only for RAID 1 (or 0), all other RAID levels require more disks and apparently the server can only accommodate two disks. RAID 10 requires four disks.
At this point in time I believe that a controller upgrade with write cache enabled (with BBU) could make the difference.
I do not see any memory issues (4GB) or CPU load (5%).
Or they have to upgrade the complete server as an alternative step.
Have you tried to copy a 5GB file from one directory to another one on the same disk (RAID 1 volume)?
Update: I ebayed a PERC 5/i with BBU and will see if it works and if it makes the difference.
Xwpis ONOMA
7 Posts
0
February 8th, 2012 12:00
Any chance you can point me to a working link for the location of that LSIUtil that enables the disk write caching? Thank you very much!
JOHNADCO
847 Posts
0
February 9th, 2012 07:00
What O/S?
Xwpis ONOMA
7 Posts
0
February 9th, 2012 09:00
Thanks for the fast response and my apologies for not mentioning that I am using Windows Small Business Server 2003 (32 bit).
I have also, duplicated the problem with Windows XP Professional w/SP3 and Windows Server R2 Standard w/SP2 (both 32 and 64 bit) and it is exactly the same. Very poor write performance (e.g. ~ 57 Mbps) and good read performance (e.g. ~ 600Mbps). I am not an IT (tech savvy) person but my understanding is that for a server (in this case a storage/backup server) anything below -say- 300Mbps is considered slow with ~500Mbps being the minimum?
Please correct me anywhere I am wrong, short answer is I am going to use this SAS 5/iR in a Windows Server 2003 environment.
Thank you very mcuh in advance!
JOHNADCO
847 Posts
0
February 9th, 2012 12:00
So in device manager, you right click on the drive and can't find a check box that says "enable write cache on the device"? or something like that? I'm a little weak on 2003 server these days.
Xwpis ONOMA
7 Posts
0
February 9th, 2012 14:00
No, because there is an add-in controller present and a RAID-1 array configured so the Windows Device Manager under 'Drives' does not display individual drives any more but a "drive device". If I right-click this and go under properties the enable Disk Cache box is not enabled (grayed out). What I am looking for is the link for the LSI utility that will allow me to set this from inside windows. Thanks!
powerpodge
1 Rookie
1 Rookie
•
87 Posts
0
January 26th, 2013 14:00
The standard LSI MegaRaid utility allows you to set the disk cache on/off if you right click the Virtual Disk in the management page.
The version of the MegaRAID utility I used was under Support and Downloads at:
http://www.lsi.com/products/storagecomponents/Pages/LSISAS1068.aspx