zfs very poor performance compared to ufs due to lack of cache?

[jhell]

On 09/16/2010 02:35, Andriy Gapon wrote:
> on 16/09/2010 00:54 jhell said the following:
>> On 09/15/2010 17:30, Steven Hartland wrote:
>>> ----- Original Message ----- From: "jhell" jhell at DataIX.net
>>>
>>>> On 09/15/2010 06:54, Steven Hartland wrote:
>>>>> ----- Original Message ----- From: "Jeremy Chadwick"
>>>>> <freebsd at jdc.parodius.com>
>>>>>> Looks like Andriy just committed something to HEAD/CURRENT which might
>>>>>> address this:
>>>>>> http://www.freebsd.org/cgi/query-pr.cgi?pr=kern/141305
>>>>>
>>>>> Already running that as part of the patches unfortunately, it doesn't
>>>>> seem
>>>>> to have any effect.
>>>>>
>>>>
>>>> Is it ? vm_page_set_validclean(m, off, bytes);
>>>> I recall you saying that you added this from earlier in the thread.
>>>> could be wrong though but what Andriy committed was the following.
>>>>
>>>> or ? vm_page_set_valid(m, off, bytes);
>>>
>>>
>>> Ahh good catch I have:
>>>    if (error == 0)
>>>        vm_page_set_validclean(m, off, bytes);
>>>
>>> and not as mentioned by 141305:
>>>    if (error == 0)
>>>        vm_page_set_valid(m, off, bytes);
>>>
>>> Which should it actaully be?
>>>
>>
>> Looking at the manual page vm_page_bits(9) I don't see a
>> vm_page_is_validclean so really would it have a effect ?.
>>
>>
> 
> Maybe the man page doesn't have it, but the function is real :-)
> In this case it actually doesn't matter much which one to use, but what was
> committed is more correct (as you might have expected).
> 

Yeah that's what I thought since the data is clean in the first place
that extra ability to zero off the end bits shouldn't ever need to happen.

Notice though I mixed up vm_page_set* with vm_page_is*, I must have been
sleeping during that point ;).

-- 

 jhell,v