Re: [Scst-devel] Some performance questions

[Vladislav B. <vs...@vl...>]

Ruben Laban, on 10/30/2009 01:29 PM wrote:
> We currently use SCST on a wide variety of hardware (ranging from dual quad 
> core systems with high-end adaptec raid cards holding 16 sas disks to single 
> core systems with just 2 sata disks in software raid). I was wondering in 
> which scenarios the various kernel patches would yield the biggest 
> differences. Currently we use stock Ubuntu kernels (2.6.24 and 2.6.28 based) 
> without any SCST patches. I'd prefer sticking to stock kernels, but if the 
> benefits would be big enough, rolling our own kernels could become worthwhile 
> the effort.

The biggest difference is one or more initiator accessing simultaneously 
data from the same real HDD-based device on the target. The biggest 
difference I've seen is 7 times. Search LKML archive for my tests and 
results.

> A slightly related question is about the various kernel versions. There have 
> been several performance comparisons between kernel versions in the past, but 
> I can't seem to find any recent ones. Am I just looking in the wrong places 
> or hasn't any bothered to run any comparison tests lately (including posting 
> them for the public to see).

I neither know any such places. But generally the storage subsystem in 
Linux is constantly improving, so newer kernel most always means better 
kernel. But you need a bit more sophisticated workloads than a plain dd 
or IOMeter to see the difference. Although sometimes it can be noticed 
even on them.

Vlad