Why are inserts so slow?

Well, inserts visit more nodes. This means:
a) They take longer, (quite a lot longer) and
b) They are more likely to get a RejectedOverload (or a timeout).

Since our calculation of when we can send inserts is based solely on
these two factors... this results in us sending inserts significantly
less frequently than requests.

What can we do about it?

Well, firstly, is it justified? It's the logical application of the load
balancing algorithm we concocted, which is based on TCP/IP and therefore
reasonably credible. Having said that, aforesaid algorithm is based on
fatal failures; packet collisions. These are not split into two
categories, and there are not more chances of one happening for a
particular kind of packet (well I suppose larger packets...)

Maybe we should use the insert round trip time, but the request (or
overall) probability? One problem with the former is that we may not
have enough requests during bootstrapping; this can be fixed when we add
automatic verification of inserts...
-- 
Matthew J Toseland - toad at amphibian.dyndns.org
Freenet Project Official Codemonkey - http://freenetproject.org/
ICTHUS - Nothing is impossible. Our Boss says so.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 189 bytes
Desc: Digital signature
URL: 
<https://emu.freenetproject.org/pipermail/devl/attachments/20060411/47fc9d46/attachment.pgp>

Reply via email to