|
Hi,
I am sorry for not mentioning the following in my
earlier mail:
The metrics for every LVS would ofcourse have to be periodically
recomputed.
However the reason I mailed again was because I had this sudden flash of
insight:
Suppose a site has more than one POP(supersparrow terminology).
Every POP is an LVS.
Every LVS Director is capable of the following:
-Sending pseudo requests to all other POPs and hence deriving the requisite
metric for that POP.
-Based on this metric determining whether or not it is the best POP to handle
the request.In case it is, then the request is forwarded to a local
realserver.
In case it isn't it redirects it to the optimum POP.
This requires that all the POPs be known initially and that newer POPs be
added manually.However failed POPs would have appropriate metrics and hence
requests wouldn't be directed to them.
This also requires that the POPs go out of action for some time when the
metric computations are being performed.This can be handled by seeing to it
that all the POPs perform them at different times.
A mechanism for avoiding infinite looping would also be required.
The advantage here is that the master director is not required.
I suggest that initial POP selection is done on a DNS RR basis.
The capability for redirection at each POP would annul the reduced
granularity due to DNS RR.
Again Please correct me if my logic is faulty.
ciao
abhijeet
|
