I know some of it is likely to be recently introduced transport layer bugs, 
and I am trying to chase them down. However, a possible reason for the recent 
low payload percentages may be that we are doing more SSK requests, because 
of the increasingly widespread use of FMS, and because those who still use 
Frost download massive quantities of invisible and useless spam messages?

If this is true, it will likely get worse over time.

The solution, which we cannot implement until after 0.7.0 has shipped, 
especially as bugs in it could be extremely disruptive, is to implement the 
RecentlyFailed mechanism in ULPRs, which will limit the number of requests 
for the same key that can happen within a certain period of time, sending a 
(fatal) RecentlyFailed error with a timeout to too eager requestors (who will 
be notified anyway if it is found because of ULPRs). This is similar to the 
old 0.5 failure tables system (which was also developed to deal with 
polling), but will work better because of being combined with ULPRs and not 
being able to self-seed (which on 0.5 was catastrophic).

One worry is that we may need to track more keys than will easily fit in 
RAM...
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: not available
URL: 
<https://emu.freenetproject.org/pipermail/devl/attachments/20080306/37489e39/attachment.pgp>

Reply via email to