Dimitri,
I've written quite a bunch on this already, and you are free to borrow
from any of the NERD versions you would like to snip from.
HOWEVER...
The title of the draft explicitly states "operational experience with
cache based mapping". Can you please help me with that? I would think
that as far as such mappings are concerned there are really only two
groups of people who have any: LISPers and HIPsters.
If we are just going to talk about previous uses of caching, that's okay
too, but I would suggest that the impact of a cache miss is HIGHLY
variable based on application, and the actual occurrence of a cache miss
is highly variable not only based on application, but based on
configuration as well.
As a case in point, I used netnews to distribute the precursor of the
Human Genome Project (GenBank) throughout the net to scientists' systems
who would then suck the thing into an RDB. Consistent distribution was
important, and so a cache miss would stall an update. But once the file
had been loaded, any further misses were irrelevant. There are a lot of
netnews design examples out there from the server side.
It's been said that we've done caching before on routers, and this is
true. And so my point is simply this: the question must be asked, what
are the similarities and differences between the systems in use and what
was done previously?
Eliot
_______________________________________________
rrg mailing list
[email protected]
https://www.irtf.org/mailman/listinfo/rrg
