Please help a noob

[UselessUser]

Hi all,

I have changed jobs and been thrown into a awful mess of an environment. I have been tasked with looking at backups (Which are god awful)..

Anyways I have downloaded the HP L&TT software and ran a System Performance test on a server which has a volume from the SAN which is showing throughput of 16mbps??? I am assuming it should be much much quicker?

The SAN consists of 2 MSA1000's in an Active Passive MSA1500cs shelf. There are two HP 2/8 san switches and there are 6 servers each with 2 Emulex 2GB Fibre cards (1 in each switch). The MSA's are connected to a MSA20, and a MSA 30, one filled with SAS 250GB disks, the other filled with SCSI 146GB 10K disks. The lights on the MSA's indicate that the load is minimal (All three LED's are on the majority of the time). The SAN switches show the ports are being utilized as F ports indicating that the SAN is switched as opposed to FC-AL...

I am new to SAN's so would appreciate if anyone could give me any starting points and a rough idea of the speed I should be expecting?

[Bill Costigan]

reading and writing to a disk array is very different. When reading, you have to look at both the read requests and amount of data requested in each read. Use perfmon to collect these statistics.

The array stripes data across all the spindles. Let's say each disk gets 64KB blocks. If you randomly read 64KB blocks the array must retreive this from a single disk. The I/O will be at the speed of a single disk. Let's assume a single disk can do 200 I/Os per second. This would result in an I/O rate of 64KB * 200 /second or 12800KB or 12.8MB/s. If you do 128KB reads, the array will read two blocks from 2 disks at the same time. each disk will still only do 200 I/Os per second but the result will be 25MB/s. The bigger the read size the more bandwidth. But as long as you need to go to a disk you will not get more than 200 I/Os per second.

If you read sequential blocks some arrays will read ahead and have the block in cache when it is requested. Under ideal conditions, this could result in I/O rates equal to 200 I/Os /sec times the number of disks. However, most backup agents don't read like this. It requests reads randomly across the disks and the array cannot predict what block will be requested. The result is read rates equal to a single disk.

Using perfmon statistics you can see the number of reads/sec. Bytes/read and disk response time. If the response time is less than 2ms the data is coming from cache. Over 2ms means the array had to hit the disk.