[indiana-discuss] terrible lofi performance?

[Andrew Gabriel]

Fajar A. Nugraha wrote:
> On Tue, Mar 24, 2009 at 4:01 PM, Fajar A. Nugraha <fajar at fajar.net> wrote:
>   
>> On Tue, Mar 24, 2009 at 3:48 PM, Andrew Gabriel
>> <agabriel at opensolaris.org> wrote:
>>     
>>>> 100% busy on 147 kbps read? Seriously?
>>>> The ISO file is located on zfs pool on sd2
>>>>
>>>> Is this expected, or is it something wrong with my setup?
>>>>         
>>> I suspect it's bugid 6806627
>>> http://bugs.opensolaris.org/view_bug.do?bug_id=6806627
>>>       
>> Ouch. So does that mean zfs is pretty much unusable (performance wise)
>> in snv_109-snv_111? Or is it one of those rarely-happened case where
>> mounting ISO happens to be one of them?
>>     

I don't know the precise circumstances which trigger it. I have not 
noticed ZFS performance to be generally bad on snv_109 which is what I 
currently have on my home server (but I might not have pushed it hard). 
It's been noticed with lofi in particular.

> After reading the bug report more thoroughly, it seems that the bug
> triggered when there's mismatch in block size and request size. So I
> tried setting recordsize=512, and copy the iso again (so it use the
> new block size). Here's what it's at now:
>
> device      r/s    w/s   kr/s   kw/s wait actv  svc_t  %w  %b
> lofi1     565.7    0.0 4513.8    0.0  0.0  0.9    1.6   1  92
> ramdisk1    0.0    0.0    0.0    0.0  0.0  0.0    0.0   0   0
> sd0         0.0    0.0    0.0    0.0  0.0  0.0    0.0   0   0
> sd1         0.0    0.0    0.0    0.0  0.0  0.0    0.0   0   0
> sd2      7696.2    0.0 7849.8    0.0  0.2  1.8    0.3  16  71
> sd3         0.0    0.0    0.0    0.0  0.0  0.0    0.0   0   0
>
> seems like a usable workaround :)

Thanks, I might add that to the bugid. Is that the ZFS recordsize you 
set (in which case, you presumably copied the iso into a new file 
afterwards in order for the file to pickup the new recordsize?)

-- 
Andrew