On Sun, 1 Dec 2013, James Harper wrote:

> I'm investigating a performance issue, and vmstat is showing:
>
> procs -----------memory---------- ---swap-- -----io---- -system-- ----cpu----
>  r  b   swpd   free   buff  cache   si   so    bi    bo   in   cs us sy id wa
>  4  3 846000 162048   1220 3002540  596    0 70008   766 18552 15943  3  6 67 
> 24
>  2  3 845824 159284   1220 3008828 1116    0 73108 17183 17748 14786  6 11 54 
> 28
>  0  4 845816 212872   1220 2958544    4    0 51452 57134 19235 16206  4 10 58 
> 27
>  6  1 845812 156088   1220 3010976    4    0 65652   991 15372 13947  1  3 75 
> 20
>  0  0 845808 167580   1220 3006876    0    0 63668 12689 20513 13451  2  4 84 
>  9
>  1  0 845808 163356   1220 3011124    0    0 32868    80 11735 12122  1  2 93 
>  4
>  0  1 845788 161248   1228 3008176   20    0 41712   540 13442 14301  1  3 88 
>  8
>  1  0 845780 162364   1228 3012860   16    0 70844 13689 20596 16787  4  6 79 
> 12
>  2  2 845756 162736   1228 3013604  112    0 36804 32394 16490 16794  1  4 78 
> 16
>  3  1 845752 159280   1228 3009800   24    0 82608   890 17589 14989  2  5 69 
> 23
>  0  3 845688 161016   1228 3017948  104    0 90356 25875 24452 15321  3  8 72 
> 16
>  1  3 845580 163248   1228 3008712  208    0 84528 12806 20642 14609  3  5 71 
> 20
>  0  1 845516 160536   1228 3019424   56    0 60240 16853 16595 13174  2  4 78 
> 16
>  1  0 845444 166364   1228 3012828   76    0 22652  1674 12705 12765  1  2 91 
>  6
>
> There is activity in the si column, and iostat on the swap device is showing 
> some activity. I'm wondering if 'si' also includes activity on memory mapped 
> files, or if it is purely pagefile access?
>
> And is a figure like 100 high? I think it reflects 100kb so is tiny, and 
> probably not significant.

100 is indeed 100kB/s, and is tiny.  drives capable of 70 or so IOPs seem
to be able to get about 3000kB (and it's been annoyingly at this level for
a decade or so for me).

You would almost expect "swap in" to include mmapped data, wouldn't you -
it's the same part of the kernel surely?  But in this case, unless you're
also calling msync() (apt-get does - and there's a current open bug
against zfs because msync is pessimistically crap currently, but there is
a patch in the works), then the rates for mmap can't possibly be that low.
70-90MByte/s is about as good as you'll get in real life on everything but
SAN hardware with 8gbps fibrechannel.

-- 
Tim Connors
_______________________________________________
luv-main mailing list
[email protected]
http://lists.luv.asn.au/listinfo/luv-main

Reply via email to