[dtrace-discuss] cleanrate?

[Jon Haslam]

Well, my excuse it's that you can only do so much trap code
in  a day before you go mad...

The comment I made below about tunables that specify a frequency
is (mostly) bogus. By specifying the unit of 'hz' you get what you ask
for. So:

#pragma D option cleanrate=101hz

does the same thing as:

#pragma D option cleanrate=9900990

I think the docs could be made clearer though. I'll go back to
my happy place now...

Jon.

> Hi Jim,
>
> If you are seeing messages of the form "dtrace: 1234 speculative drops"
> then you should (usually) try pushing the 'specsize' variable up to
> increase the size of the speculative buffers - I think it's 32k by 
> default.
> 'nspec' is used to control the number of buffers.
>
> Whilst we are here, just a quick note on the cleaning rate tunable,
> 'cleanrate'. All the tunables which specify a frequency (cleanrate,
> aggrate etc.) are actually assigned a value which when used as the
> divisor of 1000000000 will give the desired frequency. Therefore to get
> 101hz the default value for cleanrate is 9900990 and you therefore need
> to tune this down to increase the default frequency! I know this isn't
> what the docs say w.r.t the cleanrate tunable.
>
> Yes, this isn't actually user friendly. I don't know why it was done 
> this way -
> maybe one of the guys can comment on this.
>
> Jon.
>
>> I'm trying to avoid some drops from speculating and was trying to 
>> manipulate
>> cleanrate to change the rate at which speculative buffers are 
>> processed (at least
>> that's what I think it does).
>>
>> No matter what I set cleanrate to (even the default value of 101)
>> I see the message: dtrace: cleaning rate reduced to 5000 hz.
>>
>> I am speculating at a fairly high rate and see lots of drops but if I 
>> push nspec up too far, I see memory allocation failures or induce
>> undesired side effects by consuming lots of system memory.
>>
>> Suggestions?
>> -- 
>> This message posted from opensolaris.org
>> _______________________________________________
>> dtrace-discuss mailing list
>> dtrace-discuss at opensolaris.org
>>   
>
> _______________________________________________
> dtrace-discuss mailing list
> dtrace-discuss at opensolaris.org