Could it be that a 5 year old Sun T3 is 2x faster than a new Sun 3320?

Question

We just purchased a brand new system to replace a system that has been in production for over 5 years and I am finding very disappointing performance results.
The old system is a SunBlade 2000( 2x 900 MHz, 4 GB RAM, Solaris 9) with a Sun T3 fiber array (9 x 36 GB 10K RPM drives, RAID 5 with a hot spare).
The new system is a Sun T2000 ( 8-Core 1.2 GHz, 16 GB RAM, Solaris 10) with a Sun 3320 SCSI array (12 x 300 GB 10K RPM drives, RAID 5 with a hot spare) + Ultra 320 SCSI card.
I first moved over our applications and ran a few tests and found that jobs were taking about 50% longer to complete. So I decided to take the application out of the equation and just run some basic test to compare the 2 systems.
Using some basic dd tests, I could push about 90 MB/s through the T3 and only about 40 MB/s through the 3320. I also tried running IOZone and it showed the same results. The T3 was 40%-50% faster on all reads, writes and combination operations. I tried all sorts of configurations with the 3320 including single bus and split bus as well as different RAID levels. Nothing seems to help the 3320.
I've opened a support case with Sun but they are bouncing me around support from group to group, from hardware to storage to kernel and back again. My VAR is doesn�t seem too interested in helping either. I'm still hoping for the best because my upper management is not happy that we spent over $65K for new equipment and our applications are going to be slower.
Could it be that a 5 year old Sun T3 is 2x faster than a new Sun 3320?

Answer 1

Not sure if any of this will help. but it doesn't hurt to cover the basics. You might want to start by double checking the SCSI negotiation between the server and the array. Below is an example from our 3310 that we've deliberately misconfigured.
sccli> show channels
Ch Type Media Speed Width PID / SID
0 Drive SCSI 80M Wide 6 / 7
*1 Host SCSI ASYNC Narrow 1 / N/A*
2 Drive SCSI 80M Wide 6 / 7
*3 Host SCSI ASYNC Narrow N/A / 1*
6 Drive FC(L) 1G Narrow N/A / N/A
7 Host LAN N/A Serial N/A / N/A
sccli>
What we have here are host channels that have not negotiated UP to desired parameters. Please keep in mind that we've set this up purposely in our lab for training. As you can see the host connections never go into synchronous transfer and they also never negotiate to a wide bus width. This effectively throttles down the connection between the server and array. There are several causes for these types of symptoms.
- Faulty or incorrect termination.
- Mismatched hardware.
- Faulty cable
- HBA drivers
- Incorrect SCSI settings.
Since you mentioned that you have Ultra 320 HBAs and the array is capable of negotiating up to these speeds, I'd suggest you check the negotiated link speed between your array and server. If you find that the SCSI channel is not negotiating up to the desired value, one place you can check would be the "SCSI Options" in the /etc/system file.
Below is a bit mask for the various setting options
* SCSI subsystem options
* Following are applicable to all interconnects
#define SCSI_OPTIONS_LINK 0x10 /* Global linked commands */
#define SCSI_OPTIONS_TAG 0x80 /* Global tagged command support */
* Following are for parallel SCSI only
#define SCSI_OPTIONS_DR 0x8 /* Global disconnect/reconnect */
#define SCSI_OPTIONS_SYNC 0x20 /* Global synchronous xfer capability */
#define SCSI_OPTIONS_PARITY 0x40 /* Global parity support */
#define SCSI_OPTIONS_FAST 0x100 /* Global FAST scsi support */
#define SCSI_OPTIONS_WIDE 0x200 /* Global WIDE scsi support */
#define SCSI_OPTIONS_FAST20 0x400 /* Global FAST20 scsi support */
#define SCSI_OPTIONS_FAST40 0x800 /* Global FAST40 scsi support */
#define SCSI_OPTIONS_FAST80 0x1000 /* Global FAST80 scsi support */
#define SCSI_OPTIONS_FAST160 0x2000 /* Global FAST160 scsi support */
#define SCSI_OPTIONS_FAST320 0x4000 /* Global FAST320 scsi support */
Most systems have a setting of 7f8 which would only bring you to Ultra Fast Wide at 40MB per sec. Factor in the wide bus and your effective through put would be 80MB. If you find that the SCSI Options on your system are not set to support the Ultra 320 HBAs, you may want to bump up the settings here.
On the array side, you could also check to see if write cache is turned on and working.
sccli> show cache-parameters
mode: write-back
optimization: sequential
sync-period: disabled
current-global-write-policy: write-through
sccli>
In this array, the global cache setting is correctly set for write-back, but because of a fault in the array, the cache policy has defaulted to write-through. This is most common in single controller arrays (which this is). The array requires two operational controllers for write cache to be in effect.
Hopefully there is something here that you can use....