[mnet-devel] notes from IRC -- who to hello

[Bryce Wilcox-O'Hearn]

> thanks, i had read that code already. it leads to not asking 
> known-to-be-online MTs.

How can that happen?

The MT which has failed the least recently is always returned by peerman.  In 
addition, other MTs in the same "liveness level" are also returned.  So for 
example if there are two MTs in liveness-level-0, and then one fails (i.e., you 
can't open a TCP connection to it, or you send a message to it and then you 
don't hear back within the timeout), then the one that failed will be demoted to 
liveness-level-1 and (with 50% probability) will be excluded from the next 
hello.  However, if the other one also fails, then both will be demoted to 
liveness-level-1, and since liveness-level-0 will then be empty, all MTs will 
get promoted one level until liveness-level-0 is non-empty again.

I don't see how the current algorithm can lead to a problem where there is an 
MT which is known to be online but which is not helloed.

Oh, but maybe it is the log(n) selection that is causing the problem.  For 
example, if there are 3 MTs returned from peerman, then this algorithm sends 
"hello" to only two of them.  You are saying that this "minimum MTs" number 
should be not "mojoutil.logish()", but a higher number?  That sounds reasonable.

But that there is already a mechanism in place to handle the case that all of 
the hellos fail.  In that case, it demotes the failing MTs and tries again, 
I *think*.

Before increasing the "minimum to hello" number we should see if that "retry 
hello on failure" mechanism didn't work, and if it didn't work why not.  Do you 
agree?

Regards,

Zooko



-------------------------------------------------------
This SF.net email is sponsored by: Perforce Software.
Perforce is the Fast Software Configuration Management System offering
advanced branching capabilities and atomic changes on 50+ platforms.
Free Eval! http://www.perforce.com/perforce/loadprog.html
_______________________________________________
mnet-devel mailing list
mnet-devel at lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/mnet-devel