Hi.
It should speed-up route table update on new received prefixes/peer
failure. IMHO it's mostly acceptable for a big excahnge points with a
lot of peers, or - for small embedded systems (or even routers
controlplane) that work as BGP routers.
On generic soft-routers speedup will be not noticeable (for ex.,
receiving FV at less than 1 second instead of 5-10 - it's a good profit,
but it happens once per long period).
15.01.2015 00:14, Baptiste Jonglez пишет:
Hi,
Using a bloom filter certainly looks like a good idea to get a different
memory/speed tradeoff.
However, I'm a bit curious: in which situation would Bird need fast lookup
of a particular prefix in its entire FIB? I guess the « show route for
X.X.X.X » command would benefit from this optimisation, but that's not
exactly the day-to-day routine of a Bird router. Unless this might be
useful in filters?
Thanks,
Baptiste
On Fri, Jan 09, 2015 at 06:06:23PM +0000, Bahram BahramBeigy wrote:
We have accelerated FIB lookups (fib_find() and fib_route()) using a data
structure named Bloom Filter when number of inserted nodes into FIB becomes
huge, for example more than 200,000 IPs are inserted into one FIB.
Consequently, the linked list chains becomes huge, the Bloom filter avoids
traverse these long chains when an IP cannot be found.
The paper:
http://ieeexplore.ieee.org/xpl/articleDetails.jsp?arnumber=6999678
Please let me know your opinion about it.
Or any suggestions ...
Thanks in advance.
