[Users] poor performance inside a VE

[Ian P. Christian]

Hi all,

A friend recently wanted to do some rendering, so I setup him up a VPS 
quickly on an idle machine for him to render on.... however - it didn't 
go so well.

Here's the output of dstat with the rendering goign on in VE0 (the host).


----total-cpu-usage---- -disk/total -net/total- ---paging-- ---system--
usr sys idl wai hiq siq|_read write|_recv _send|__in_ _out_|_int_ _csw_
  17   1  82   0   0   0|  22k  637k|   0     0 |   0     0 | 236   948
  54   0  46   0   0   0|   0     0 |2773B 4860B|   0     0 | 126   818
  61   3  36   0   0   0|   0    16k| 596B  360B|   0     0 | 141   759
  77   1  22   0   0   0|   0     0 |2835B 4840B|   0     0 | 118   707
  81   1  18   0   0   0|   0    80k|   0     0 |   0     0 | 124   747


As you can see, it's using most of the CPU (there's 2 hyperthreading 
xeons in this machine), and all is well. The render completes in about 
30 minutes.


However... here's what happens when I run ths render inside a VE
----total-cpu-usage---- -disk/total -net/total- ---paging-- ---system--
usr sys idl wai hiq siq|_read write|_recv _send|__in_ _out_|_int_ _csw_
  18   1  81   0   0   0|  22k  626k|   0     0 |   0     0 | 234   971
   2   0  98   0   0   0|   0     0 |2675B 4500B|   0     0 | 120    11k
   0   0  99   0   0   0|   0    16k|   0     0 |   0     0 | 126    12k
   2   0  98   0   0   0|   0     0 |1999B 3730B|   0     0 | 121  9010
   1   0  99   0   0   0|   0    16k|   0     0 |   0     0 | 125    11k
   1   0  99   0   0   0|   0     0 |1935B 3730B|   0     0 | 116    10k


Notice the stupidly high number of context switches (last column).  The 
render on the VE is estimated to take about 10 times longer...

# in=spaceinvader.flam3 format=png transparency=false prefix="renderc" 
verbose=1 flam3-render
Automatically detected 4 core(s)...
chaos:   0.5%  ETA: 286.5 minutes


Can anyone suggest why this is?


-- 
Ian P. Christian ~ http://pookey.co.uk