FAS and V-Series Storage Systems Discussions

Highlighted

Re: CPU High Usage On my filer FAS2040 A-A

Hi,

I agree with Radek; this system is simply near bottleneck.

Take a look at this snippet:

Summary Statistics (   70 samples  1 secs/sample)

CPU   Total    Net kB/s    Disk kB/s    Tape kB/s Cache Cache  CP  CP Disk

       ops/s    in   out   read  write  read write   age   hit time ty util

Min

69%     683  7947 140887 156703      0     0     0     2s  62%   0%  *  39%

Avg

82%     861 16322 182756 190607  18342     0     0    20s  74%  59%  *  78%

Max

91%    1065 24991 210855 218952  50551     0     0    32s  81% 100%  * 100%

From the Average row:

82% CPU: Quite high.  Checking the sysstat -M output it looks balanced around processing functions

183 MB/s reads + 16 MB/s writes = 199 MB/s network xfer: You are using at least 2 x GbE links.  You might hit link bottleneck if traffic is not balanced.

861 IOPs:   take the 199MB/s / 861 IOPs = 231 KB / op.  So you are doing large IOs. 

78% disk busy: Above 50% and you can already start to see some variability in latency.  At 78% average the disks are quite busy; they are processing at least one xfer 78% of the time so likely some IOs are also waiting to be processed at times.  If IOs are waiting at the disk layer these will eventually make IOs wait on the host side.

Disk IO efficiency: Disk read is 190MB/s, network out is 182 MB/s which is roughly balanced.  Disk write is 18MB/s, network in is 16MB/s which is roughly balanced; the extra little bit is parity.  You have no evidence of freespace fragmentation.

Reallocate is useful if you have freespace fragmentation (you don't) or if you have sequential read after random write (you seem to be doing large block IO so I doubt it).  Even then, it is has a bigger impact on reducing disk busy than cpu busy.  Checking your disk IO efficiency however I'd say it looks pretty good; network in/out matches disk in/out.

So across the main system aspects (CPU, Disk IO, Network) on each you are nearing bottleneck.  The solution is to stay as-is and be satisfied with latency you have, reduce workload (or move to the other head if it is less busy) and see latency improve, or buy a more powerful controller.

Hope that helps!

Christopher Madden

Storage Architect, NetApp EMEA

Highlighted

Re: CPU High Usage On my filer FAS2040 A-A

Hi Radek,

Thanks a lot for making me to understand things better hope will run the reallocate and will see for the optimization value.

Madden,

Thank you so much for your crystal clear explanation about the busy stats from the output above. We are looking forward to buy a High end filer and migrate all our stuff to the new one .

And also I am confused whose response i need to mark as CORRECT ANSWER so that it will be helpful for others .

Thanks,

Nayab

Highlighted

Re: CPU High Usage On my filer FAS2040 A-A

Fragmentation / reallocation is a bit of a side-tracking really, as it is fairly obvious your system is under heavy load - so I suggest you mark one of the answers mentioning this as correct.

(I actually think when people search communities, they tend to read a whole thread anyway, rather than looking only at the "correct" answer without its context)

Check out the KB!
NetApp Insights To Action
All Community Forums