Title: | HSD30 Product Conference |
Moderator: | SSDEVO::EDMONDS N |
Created: | Mon Apr 11 1994 |
Last Modified: | Tue Jun 03 1997 |
Last Successful Update: | Fri Jun 06 1997 |
Number of topics: | 300 |
Total number of notes: | 1008 |
I am trying to reduce response times on a number of disks. The disks currently hang off HSD30 controllers with 32Mb cache. Decps shows queues building up on the disks, so I am investigating controller based striping. Due to space limitations I can only connect 2 shelves to the controller. This presents me with a dilema. Ideally I would have 3 member stripesets. One disk on each port, but with only 2 ports available, do I drop down to 2 member stripsets (each on a seperate port) or have a 3 member set with 2 disks on the same port ? Which is likely to give better response times ? Combined peak KB thruput to the existing disks is 1395, but the average is around 400 KBytes. Peak combined queue length is 7.5, mainly (6) on one of the disks, hence the reason for striping. Any thoughts ? Andy Williams OMS London
T.R | Title | User | Personal Name | Date | Lines |
---|---|---|---|---|---|
293.1 | KERNEL::LOANE | Comfortably numb!! | Thu Mar 20 1997 13:49 | 3 | |
You could start by putting one of the BA350's into split bus mode (at least this would give you access to all 3 SCSI ports). Have you done any analysis with VTDPY?? | |||||
293.2 | POMPY::WILLIAMS | Fri Mar 21 1997 04:07 | 16 | ||
> You could start by putting one of the BA350's into split bus mode > (at least this would give you access to all 3 SCSI ports). Thanks, Great suggestion. I think split bus mode is the answer. > Have you done any analysis with VTDPY?? No, I've just used the DECPS data, what other analysis do you think needs performing ? Andy | |||||
293.3 | GEM::SHERGOLD | We are 100% sure; well almost!! | Fri Apr 18 1997 13:00 | 4 | |
It may seem a silly question but you do have cache turned on on these units don't you? A lot of people think that caching is automatic. Keith |