[bind10-dev] performance issue about view
Jeremy C. Reed
jreed at isc.org
Mon Jul 20 11:40:39 UTC 2009
On Mon, 20 Jul 2009, xinpu wang wrote:
> I?m not sure whether this mail is proper here, so put my apology first.
>
>
>
> I have been using BIND as a CDN DNS server for a while, and found that the
> performance would be significantly affected if the numbers of views and ACL
> rules of each view are large.
>
>
>
> In bind-9.3.4, to find the match view for a client IP, two lists are need to
> be searched, one list for views and each view has a list of ACL rules. Under
> this condition, the performance is extremely poor when I tested.
>
>
>
> IN bind-9.6.0 I used, the list of ACL rules is replaced by a tree, so the
> number of ACL rules impacts the performance little. But views are still
> stored in a list, and if there are many views, it also needs much time to
> search every view from head to tail.
>
>
>
> I made a little change, simple and ugly, using an array to store the mapping
> relationship between IP and the view it can access. This can increase the
> performance from 17%to 100%(the number of views from 10 to 100).
>
>
>
> I?m wondering whether this issue deserved a more completely consideration
> and an elegant design, while I aware that the aim of BIND maybe is just a
> general function DNS server and adding ten views with tens of thousands ACL
> rules sounds not a common situation.
>
>
>
> Thank you for your time.
This list is for BIND 10 development and planning. We will make a note of
this issue in regards to BIND 10 and I may setup some benchmarking
performance testing for many views with many ACLs. (I have already done
many ACLs and have seen the great improvement after 9.4.x but I didn't
benchmark the views performance for that.)
Please consider sharing your patch to bind9-bugs at isc.org or post to the
the bind-workers list to discuss BIND 9 development.
https://lists.isc.org/mailman/listinfo/bind-workers
If you'd like to participate in BIND 10 development, please continue to be
on this list.
Jeremy C. Reed
ISC
More information about the bind10-dev
mailing list