Re: [lisp-snmp] Misc fiddles for new OID types

[Chun T. (binghe) <bin...@gm...>]

All right, I'll see what I can do.

--binghe

在 2009-5-11，12:36， John Fremlin 写道：

> Hi Chun Tian,
>
> "Chun Tian (binghe)" <bin...@gm...> writes:
> [...]
>>> As I said before, the most sensible thing is probably just to use
>>> simple
>>> vectors as OIDs.
>>
>> I think this change wouldn't bring any essential performance boost
>> from the view of algorithm, right?
>
> If you are talking about time complexity in a naive way then that's
> true.
>
> But the time complexity in our workload is heavily affected by the GC.
>
>> A simple vector is "fast" than A list only when we need to access
>> specific elements in it. For simple vectors, we have O(1) performance
>> here (and O(n) for lists). But you should see, in ASN.1 package, any
>> time we need this vector/list, we need all elements from it. Here we
>> just waste some memory (which equals to the number of SIMPLE-OID
>> instances times a small factor), compare to the whole memory
>> requirements it's nothing.
>
> I think compared to the whole memory requirements and the strain on  
> the
> garbage collector it is very important.
>
> Why do you think that it is not important?
>
> In our profile, the biggest memory user was OID creation (instances)  
> and
> the oid-number-list creation.
>
> Both would be heavily reduced with simple vectors for OIDs
>
> [...]
>