ibm ds4300 vs ds4700

Hi
If we have 18 15k disks in a raid 10 setup for sql-server data using ibm ds4300 (without turbo).
We are currently seeing around 17,3 as avg disk queue write length during peak hours.
We dont have any read queue length.
Would an upgrade to ds4700 improve the performance, or would it be better to buy another expansion unit (exp 810) with 16 disks, so we get 18+16=34 spindles?
The difference of ds4300 and ds4700 is 4Gbps instead of 2Gbps bandwith (the hard disks that we have is 4gb disks) and more cache on the controllers.

What's the cost difference?
You'd probably see better performance from additional disk....have implemented any monitoring on the switch to see how much of the fiber bandwidth is being utilized?

[quote user="melvinlusk"]What's the cost difference?[/quote]
Let's just assume that the cost doesn't matter.
[quote user="melvinlusk"]You'd probably see better performance from additional disk....have implemented any monitoring on the switch to see how much of the fiber bandwidth is being utilized?[/quote]
We only see those write queues during checkpoints. So I was thinking a bigger cache on the SAN could make these checkpoints faster since it could write to memory (cache) and then the SAN could store it to disk. Of course I see a problem here if the SAN cache has not been able to store all the data between the checkpoints, but I guess that's not what we are discussing here.

In my opinion, it would be better to upgrade to ds4700.
However, make sure that your database server motherboard can support 4GB fibrechannel cards. You will almost certainly need a PCI-Express slot available on the board because older PCX slots do not support 4GB fibrechannel cards.

[quote user="davidfarr"]In my opinion, it would be better to upgrade to ds4700.[/quote]
Why?

[quote user="brimba"]Why?[/quote]
You have answered this yourself, in your comment below:
[quote user="brimba"]The difference of ds4300 and ds4700 is 4Gbps instead of 2Gbps bandwith (the hard disks that we have is 4gb disks) and more cache on the controllers.[/quote]
There is little benefit in having 4Gbps disks if they have to bottleneck through a 2Gbps fibrechannel. A while ago we upgraded one of our older SANs from a RAID50 SCSI array to a RAID10 SAS array (4Gbps+), but we kept the same 2Gbps fibrechannel cards because we only had PCX slots on the Intel Itanium server. The performance benefit we got from the disk upgrade was approx. 10% improvement. It was therefore very obvious to us that the 2Gbps card was an I/O bottleneck.

[quote user="davidfarr"]The performance benefit we got from the disk upgrade was approx. 10% improvement. It was therefore very obvious to us that the 2Gbps card was an I/O bottleneck.
[/quote]
Did you upgrade to 4Gbps cards? What happened to the performance after that?

avg disk queue is not good indicator for disk performance in SAN
have a look to: http://sql-server-performance.com/Community/forums/p/28525/152373.aspx#152373
http://sql-server-performance.com/Community/forums/p/29099/154329.aspx#154329

[quote user="moh_hassan20"]
avg disk queue is not good indicator for disk performance in SAN
have a look to: http://sql-server-performance.com/Community/forums/p/28525/152373.aspx#152373
http://sql-server-performance.com/Community/forums/p/29099/154329.aspx#154329
[/quote]
Ok, if I use these counters I get 6ms on avarage for read and about 27ms for write. Although write is about 0 all the time and then peak to about 35-60ms during checkpoints.
So the question still remains.
What will lover the checkpoint time the most (get better response time from the SAN).
- replace the existing ds4300 with ds4700 (or the new ds3950)
- add a new exp810 with 16 more disks

let us review SAN configuration checklist , before jumping to solution selection:

• what is the NTFS Allocation Unit Size? 64k recomended
• what is RAID strip size ?
• Is disk alignment offset and NTFS allocation unit size are set (Initially, you loose 30% performance in case of disalignment)
• what HPA Queue depth ?
• Is SAN shared with other hosts?
• are data /logs in separate LUN , and every one has its own physical disks?
• how many HBA in server?
• how many free ports available in SAN?
• What is cache size for read /write ? is it enabled?

4k as ntfs allocation unit size.
64KB
We used diskpart to align the disks to 64kb, but I understand that we should have also set the ntfs allocation unit size to 64kb.
Dont know the queue length
Yes, two other hosts (fileserver)
yes separate luns and separate drives
2 hba
4 ports available in the san switch
booth read and write cache is enabled, the size is 256Mb

[quote user="brimba"]4k as ntfs allocation unit size.[/quote]
First, you should Backup your data , and format NTFS with alolocation size 64 K
HPA Queue depth should be configured by SAN Engineer (default 32 , may be enough)
use diskpart for alignment
btw, windows 2008 resolved alignment problems

[quote user="moh_hassan20"]First, you should Backup your data , and format NTFS with alolocation size 64 K
HPA Queue depth should be configured by SAN Engineer (default 32 , may be enough)
use diskpart for alignment
btw, windows 2008 resolved alignment problems [/quote]
Thanks.