directory-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Ersin Er" <ersin...@gmail.com>
Subject Re: [ADS 2.0] New ASN.1 codec metrics
Date Sat, 09 Sep 2006 11:57:35 GMT
Hi,

On 9/9/06, Emmanuel Lecharny <elecharny@iktek.com> wrote:
> Hi,
>
> I spent some time this week to work on a new structure for codec.
> Basically, my initial intent was to merge Snickers and twix, to avoid a
> transformation. Another objective was to build a better Object hierarchy
> which will reflect the LDAP grammar given in RFC 2251 and RFC 4511.
> Last, not least, I wanted to decouple encoding and decoding from the
> objects, using decorators which can be BER, DSML or Ldif codecs.
>
> I have just finished the first step of this experiment, which consist of
> a BindRequest total recvamping, except the decoding, and here is the
> result :
> - Twix : encoding 10 000 000 BindRequest takes 27 seconds
> - NewTwix : encoding 10 000 000 BindRequest takes 192 seconds
>
> We can see that it's around 7 times slower, due to more objects being
> created, and may be too many methods calls through the Decorators. I
> have to check.
>
> However, NewTwix can encode around 52 000 BindRequest by second (don't
> be afraid : I could have done it with searchRequest, but it was too
> complicated for an experiment), which is by far excedding the actual
> server capacity (we are delivering around 900 searches response per
> second, atm). But that means we will use more CPU to encode, and right
> now, the profiling sessions I have done show that encoding represent
> around 3% of the overall server CPU consumption. I'm afraid that the new
> version may use around 20%, or even worse...
>
> Question : at what point good design is better than performance ?

Good design is good and it can be optimized I think. Don't you think
that there may be some optimization points in the new code?

> Emmanuel L├ęcharny

-- 
Ersin
Mime
View raw message