How to reduce the load on the disk subsystem in vsphere + openfiler (iscsi) RAID10?

Question

VMware

Good day to all!
I am using vshere 5.1. About 60 VMs are running on one openfiler (iscsi) iron server. It has a RAID 10 of 6 disks of 3 TB. VMs are registered on 6 esxi. Well, accordingly, openfiler is mounted to each. Connection via GB line.
The problem is that I/O is constantly loaded.

2013-12-12 10:39:49 Device:         rrqm/s   wrqm/s     r/s     w/s    rkB/s    wkB/s avgrq-sz avgqu-sz   await  svctm  %util
2013-12-12 10:39:49 sda               0.00     1.50    0.00    0.80     0.00     9.21    23.00     0.26  320.25  82.00   6.57
2013-12-12 10:39:49 sda1              0.00     0.00    0.00    0.00     0.00     0.00     0.00     0.00    0.00   0.00   0.00
2013-12-12 10:39:49 sda2              0.00     0.00    0.00    0.00     0.00     0.00     0.00     0.00    0.00   0.00   0.00
2013-12-12 10:39:49 sda3              0.00     1.50    0.00    0.80     0.00     9.21    23.00     0.26  320.25  82.00   6.57
2013-12-12 10:39:49 sdb               7.61    15.82   29.83   73.87   559.66   723.02    24.74     1.83   17.68   4.45  46.13
2013-12-12 10:39:49 sdb1              7.61    15.82   29.83   73.87   559.66   723.02    24.74     1.83   17.68   4.45  46.13
2013-12-12 10:39:49 sdc               3.90    16.82   19.12   72.57   288.84   721.82    22.04     2.76   29.91   6.99  64.05
2013-12-12 10:39:49 sdc1              3.90    16.82   19.12   72.57   288.84   721.82    22.04     2.76   29.91   6.99  64.05
2013-12-12 10:39:49 sdd               7.51    15.12   30.53   84.68   448.20   750.55    20.81     1.40   13.39   3.74  43.05
2013-12-12 10:39:49 sdd1              7.51    15.12   30.53   84.68   448.20   750.55    20.81     1.40   13.39   3.74  43.05
2013-12-12 10:39:49 sde               3.60    15.12   20.72   83.98   370.22   746.85    21.34     1.68   16.01   4.20  44.00
2013-12-12 10:39:49 sde1              3.60    15.12   20.72   83.98   370.22   746.85    21.34     1.68   16.01   4.20  44.00
2013-12-12 10:39:49 sdf               5.91    16.42   29.13   79.38   485.09   731.63    22.43     1.19   10.85   3.56  38.63
2013-12-12 10:39:49 sdf1              5.91    16.42   29.13   79.38   485.09   731.63    22.43     1.19   10.85   3.56  38.63
2013-12-12 10:39:49 sdg               2.60    15.92   21.32   80.08   330.43   732.43    20.96     0.90    8.89   3.18  32.21
2013-12-12 10:39:49 sdg1              2.60    15.92   21.32   80.08   330.43   732.43    20.96     0.90    8.89   3.18  32.21
2013-12-12 10:39:49 sdh               0.00     0.00    0.00    0.00     0.00     0.00     0.00     0.00    0.00   0.00   0.00
2013-12-12 10:39:49 sdh1              0.00     0.00    0.00    0.00     0.00     0.00     0.00     0.00    0.00   0.00   0.00
2013-12-12 10:39:49 md0               0.00     0.00  143.44  248.35  2485.99  2202.30    23.93     0.00    0.00   0.00   0.00
2013-12-12 10:39:49 dm-0              0.00     0.00  143.44  248.35  2485.99  2202.30    23.93    10.41   26.83   2.56 100.11
2013-12-12 10:39:49

Tried to experiment with local cache on esxi. So that part of the virtulok disk uses local datastores during operation. Experimented with queue parameters on esxi itself. But that didn't help much.
The goal is to "settle" 130 machines on one external datastore (openfiler).
How realistic is this and what can you advise on optimizing storage storage?
Thanks in advance.

Answer 1

Well, if your machines generate io, then you can’t reduce it.
As an option, move I / O hungry VMs to a separate datastore.
Too little information to give advice.

Answer 2

If the disks are 3 TB each, then they have no more than 7200 rpm - it may simply not be enough performance of a single screw that slows down the rest of the array.
As an option, make another storage on productive 10k-15k SAS disks and move the VMs that generate the most iops there, but it won’t be cheap ...

Answer 3

Making additional storage systems will help by itself. It is also clear that it is impossible to increase the speed of the disk. The question is how else (besides what I already did) you can configure vsphere so that it is with such characteristics of the storage system that the load decreases. Maybe you can increase the cache size with parameters or something else from this area. The question is precisely this. But thanks for the replies nonetheless.

Answer 4

try to change the LUN mapping from blockio to fileio