Try to install patch 139562-02 on the control and I/O domain.
Until that patch, I/O to a backend file are done using paging so
this generate reads when doing the paging. The patch fixes that
problem.

  Also mkfile is not a good way to evaluate performances because
it does sequential writes, and performances of virtual disks are
much better when doing parallel writes.

alex.

Luis wrote:
> I'm testing Sol 10 10/08 with Ldoms 1.1. I have 2 Domains, Control Domain and 
> a Guest Domain. Control Domain has mirrored ZFS based root and a  zvol was 
> created and exported to be used as root disk in the Guest Domain. Guest 
> DOmain boots from UFS. 
> 
> I have few interestiong observations. If in the Control Domain I run mkfile 
> to do a 1GB file I get:
> 
> # date;mkfile  1g /eraseme5.txt;date
> Wed Jan 28 23:58:58 PST 2009
> Wed Jan 28 23:59:12 PST 2009
> # 
> 
> So it only takes 14 seconds to create such file.  The output of iostat -xnz 
> 10is:
> 
>                     extended device statistics              
>     r/s    w/s   kr/s   kw/s wait actv wsvc_t asvc_t  %w  %b device
>                     extended device statistics              
>     r/s    w/s   kr/s   kw/s wait actv wsvc_t asvc_t  %w  %b device
>     0.1    0.0    0.1    0.0  0.0  0.0    0.0    4.1   0   0 c1t0d0
>     0.2    0.0    2.1    0.0  0.0  0.0    0.0    6.7   0   0 c1t1d0
>                     extended device statistics              
>     r/s    w/s   kr/s   kw/s wait actv wsvc_t asvc_t  %w  %b device
>     0.0  452.4    0.0 57123.8  0.0 24.8    0.0   54.7   2  74 c1t0d0
>     0.0  451.3    0.0 56957.4  0.0 24.9    0.0   55.2   2  75 c1t1d0
>                     extended device statistics              
>     r/s    w/s   kr/s   kw/s wait actv wsvc_t asvc_t  %w  %b device
>     0.1  310.9    0.1 38632.6  0.0 16.6    0.0   53.3   1  52 c1t0d0
>     0.0  312.0    0.0 38799.0  0.0 16.7    0.0   53.5   1  53 c1t1d0
>                     extended device statistics              
>     r/s    w/s   kr/s   kw/s wait actv wsvc_t asvc_t  %w  %b device
>     0.2    0.0    0.4    0.0  0.0  0.0    0.0    0.1   0   0 c1t0d0
> 
> 
> Please have a look that here there are only writes (as expected). 
> 
> 
> 
> When I do same operation but from guest domain I get:
> 
> # date;mkfile  1g /eraseme7.txt;date
> Thu Jan 29 00:09:26 PST 2009
> Thu Jan 29 00:10:25 PST 2009
> # 
> 
> So the operation takes 1 minute (vs. 15 seconds) . Now the iostat -xnz 5 from 
> the control domain itself looks like:
> 
>     r/s    w/s   kr/s   kw/s wait actv wsvc_t asvc_t  %w  %b device
>                     extended device statistics              
>     r/s    w/s   kr/s   kw/s wait actv wsvc_t asvc_t  %w  %b device
>     0.8    0.0   44.8    0.0  0.0  0.0    0.0    6.2   0   0 c1t0d0
>                     extended device statistics              
>     r/s    w/s   kr/s   kw/s wait actv wsvc_t asvc_t  %w  %b device
>   497.9   52.8 7873.6 6184.9  0.0  3.6    0.0    6.6   1  29 c1t0d0
>   607.6   54.9 7962.4 6479.3  0.0  3.7    0.0    5.5   2  35 c1t1d0
>                     extended device statistics              
>     r/s    w/s   kr/s   kw/s wait actv wsvc_t asvc_t  %w  %b device
>   149.1   94.1 2223.1 10932.8  0.0  5.5    0.0   22.5   1  23 c1t0d0
>   147.4   91.5 2148.7 10638.4  0.0  5.5    0.0   23.2   1  24 c1t1d0
>                     extended device statistics              
>     r/s    w/s   kr/s   kw/s wait actv wsvc_t asvc_t  %w  %b device
>   242.8  281.9 4040.9 33253.6  0.0 16.4    0.0   31.3   2  61 c1t0d0
>   225.3  283.0 3841.7 33253.6  0.0 17.9    0.0   35.3   2  65 c1t1d0
>                     extended device statistics              
>     r/s    w/s   kr/s   kw/s wait actv wsvc_t asvc_t  %w  %b device
>   421.7  156.2 6848.2 17338.3  0.0 10.4    0.0   18.0   2  44 c1t0d0
>   451.3  156.9 6908.2 17317.2  0.0 10.4    0.0   17.1   2  44 c1t1d0
>                     extended device statistics              
>     r/s    w/s   kr/s   kw/s wait actv wsvc_t asvc_t  %w  %b device
>   507.6   98.3 9587.8 11325.4  0.0  7.0    0.0   11.5   2  37 c1t0d0
>   484.2   97.3 9679.0 11346.5  0.0  9.8    0.0   16.8   1  46 c1t1d0
>                     extended device statistics              
>     r/s    w/s   kr/s   kw/s wait actv wsvc_t asvc_t  %w  %b device
>   469.3  196.2 6280.1 22790.5  0.0 11.9    0.0   17.8   2  51 c1t0d0
>   529.8  196.4 6750.5 22790.5  0.0 12.2    0.0   16.7   2  53 c1t1d0
>                     extended device statistics              
>     r/s    w/s   kr/s   kw/s wait actv wsvc_t asvc_t  %w  %b device
>     1.3    0.0   10.4    0.0  0.0  0.0    0.0    4.7   0   1 c1t0d0
>     0.2    0.0    2.6    0.0  0.0  0.0    0.0    7.8   0   0 c1t1d0
> 
> 
> 
> I wonder why in this case I see reads and writes. As you see above when I do 
> the operation from control domain there are only writes. 
> 
> The iostat -xnz 10 reported in the guest domain itself looks:
> 
>                     extended device statistics              
>     r/s    w/s   kr/s   kw/s wait actv wsvc_t asvc_t  %w  %b device
>                     extended device statistics              
>     r/s    w/s   kr/s   kw/s wait actv wsvc_t asvc_t  %w  %b device
>                     extended device statistics              
>     r/s    w/s   kr/s   kw/s wait actv wsvc_t asvc_t  %w  %b device
>     0.0   60.2    0.0 7394.3  0.0 36.8    0.0  610.5   0  38 c0d0
>                     extended device statistics              
>     r/s    w/s   kr/s   kw/s wait actv wsvc_t asvc_t  %w  %b device
>     0.0  136.2    0.0 17011.3  0.0 101.9    0.0  748.0   0 100 c0d0
>                     extended device statistics              
>     r/s    w/s   kr/s   kw/s wait actv wsvc_t asvc_t  %w  %b device
>     0.0  243.9    0.0 30075.7  0.0 85.4    0.0  350.2   0  99 c0d0
>                     extended device statistics              
>     r/s    w/s   kr/s   kw/s wait actv wsvc_t asvc_t  %w  %b device
>     0.0  141.0    0.0 16751.8  0.0 102.2    0.0  724.7   0 100 c0d0
>                     extended device statistics              
>     r/s    w/s   kr/s   kw/s wait actv wsvc_t asvc_t  %w  %b device
>     0.0   81.5    0.0 10010.4  0.0 114.2    0.0 1401.4   0 100 c0d0
>                     extended device statistics              
>     r/s    w/s   kr/s   kw/s wait actv wsvc_t asvc_t  %w  %b device
>     0.0  156.5    0.0 19381.6  0.0 94.9    0.0  606.5   0 100 c0d0
>                     extended device statistics              
>     r/s    w/s   kr/s   kw/s wait actv wsvc_t asvc_t  %w  %b device
>     0.1   42.6    0.8 4510.5  0.0 62.8    0.0 1471.1   0  63 c0d0
> # 
> 
> Here only writes are reported. However the response time looks in the high 
> side. 
> 
> Any opinion on this? 
> 
> Thanks and Regards
> Luis

Reply via email to