2

Following links collected in this thread I started to check blockdev and found the following output indicating different sizes for partition c0d9p1 and the underlying device (c0d9):

[root@machine ~]# blockdev --report /dev/cciss/c0d9
RO    RA   SSZ   BSZ   StartSec     Size    Device
rw   256   512  4096          0 3906963632  /dev/cciss/c0d9
[root@machine ~]# blockdev --report /dev/cciss/c0d9p1 
RO    RA   SSZ   BSZ   StartSec     Size    Device
rw   256   512  2048          1 3906959039  /dev/cciss/c0d9p1

We have a lot of small files, so yes the block size is smaller than normal. The device is a logical driver on an HP Smart Array P410 raid controller, simple disk without any raid - RAID 0 on one disk to be precise. (Please note that above configuration is a feature not a bug).

Therefore, I have the following questions.

  1. Can the above discrepancy in the block size affect disk performance?
  2. Can I control the block size using hpacucli?
Community
  • 1
Wawrzek
  • 202
  • 2
  • 12

1 Answers1

0

1). No, it won't affect performance. Your mounted device is the the /dev/cciss/c0d9p1 device.

2). You can't control block sizes through hpacucli. You can modify the queue depth, elevator sort, cache settings, etc. via the tool. Try hpacucli help modify. You may be more interested in lowering the stripsize.

However, on a single disk logical drive, your bottleneck will be having only one drive, versus any of the other tweaks available to you.

What are you trying to do with your performance? What is the goal?

ewwhite
  • 194,921
  • 91
  • 434
  • 799
  • Thanks for responses (and editing question -it looked as I didn't check it before sending). The goal is to find any possible performance improvement for our current cheap but not the most efficient system, without spending fortune, i.e. for building better RAID configuration. One of the line of attack is to tweak I/O elevator on the kernel level. i.e. there are suggestions that **noop** one may be better for box with raid controllers. That led me to looking more carefully on RAID configuration, blockdev output. I found some things I not fully understand and wanted to double check its meaning. – Wawrzek Sep 27 '12 at 09:34
  • @Wawrzek The `deadline` scheduler is almost always better. Also, if you are using RHEL6 or CentOS 6, try changing the [tuned-adm profile for the system, as listed in this answer](http://serverfault.com/a/431112/13325). – ewwhite Sep 27 '12 at 11:08
  • Thanks again. I'm going to give all of them a try, just need to set up test environment. – Wawrzek Sep 27 '12 at 11:58