San average speed

[cyberbootje]

Hi

Any idea what the average read/write speed in a vm is when the vm is located on a SAN?
San is connected with a Gbit switch.

I am currently getting:
Bufferd read: 71.41MB/Sec
FSYNCS/SECOND: 122.63

This is a test platform, scsi disks and Gbit connection with Gbit switch.

 

[cyberbootje]

No one?

 

[rmcewen]

I think it would vary greatly based on any number of factors, including the benchmark you're running. I ran bonnie++ on two identical vm's, one on an iscsi SAN and one on the local machine. Both the SAN and the local machine were running software raid 1.

Generic Lenny (LVM on iscsi):

Version 1.03d       ------Sequential Output------ --Sequential Input- --Random-
                    -Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks--
Machine        Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP  /sec %CP
lenny         1000M 45780  86 54308  12 43957  14 51710  96 450980  83 +++++ +++

Generic Lenny (LVM on local):

Version 1.03d       ------Sequential Output------ --Sequential Input- --Random-
                    -Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks--
Machine        Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP  /sec %CP
lostbyte       1000M 43035  69 34673   3 44721  11 63402  96 380604  88 +++++ +++

So in my case, the results were fairly comparable.

What I'd really like to see is a comparison of iscsi vs AOE.

 

[cyberbootje]

I am getting a factor 10 difference if i do a test with bonnie++ on the local disk and one on the san. That can't be correct...
I just want a average reference so i can tell that let's say, 51MB/Sec is normal or 9MB/Sec is normal on a san.

 

[rmcewen]

Understood, that makes sense. My "SAN" is just a Debian Lenny box with 4 terabyte drives running raid 1. Athlon X2 CPU with 1GB of memory. Exports nfs shares and an iscsi volume. I do have a dedicated Gbit network that the virtual hosts use to access the SAN. VM speed seems comparable to local. I ran some benchmarks with jumbo frames enabled but it didn't seem to make much difference in performance.