| Re: "strong" Disk I/O prioritisation [message #29302 is a reply to message #29281] |
Thu, 10 April 2008 10:32  |
gnutered
Messages: 12
Registered: April 2008
|
Junior Member |
|
|
Vasily, thanks.
I'm recompiling the kernel now with that commenting out you mentioned.
| Quote: |
1) When you do cat for the first time, some parts or even the whole file is in cache. So, 2nd time you do cat, it is not read from the disk, but from the main memory. It can introduce significant distortions.
|
Yes, I knew that - that's why the files are each 512MB (phys mem = 521MB too), and I never read the same one twice, and it's read sequentially. When I run the test I seem to get consistent numbers.
| Quote: |
2) In implementation that works now in OpenVZ you can notice the effect of prioritization much more, if you will run not one "disk-reader" (as in your test) in VE, but several of them. It is not the feature, but a drawback, and we're working on improvements in this area.
|
I think you're saying that my tests aren't really real-world representations, and the performance might be better on my production server, and I agree. I'm going to install a copy of the production webserver on this box, and try it then.
I look forward to future improvements in the I/O prioritisation.
I'll also get to play with the different schedulers that way too.
Tony
|
|
|
|
OpenVZ Forum
Members
Search
Help
Register
Login
Home
