在 16/8/7 21:20, Eliezer Croitoru 写道:
Hey Simon,
I do not know the plans but it will depend on couple things which can fit to one case but not the other.
The assumption that we can fetch any part of the object is the first step for any solution what so ever.
However it is not guaranteed that each request will be public.
The idea of static chunks exists for many years in many applications and in many forms and YouTube videos player uses a similar idea. Google video clients and servers uses a bytes "range" request in the url rather then in the request header.
Technically it would be possible to implement such an idea but it has it's own cost.
Eventually if the file is indeed public(what squid was designed to cache) then it might not be of a big problem.
Depends on the target sites a the solution will be different.
Before deciding on a specific solution my preferred path is to analyze the requests.
By observing amplified traffic of 500% to clients side you mean that the incoming traffic to the server is 500% compared to the output towards the clients?
If so I think that there might be a "smarter" solution then 206 range offset limit.
The old method of prefetching works pretty good in many cases. From what you describe it might have better luck then the plain "fetch everything on the wire in real time".
I cannot guarantee that prefetching is the right solution for you but I think that a case like this deserves couple eyes to understand if there is a right way to handle the situation.
I think prefetch may not be fit for forward proxy, as we do not know
what's "hot" request exactly. LRU should do more efficient.
Simon
_______________________________________________
squid-users mailing list
squid-users@xxxxxxxxxxxxxxxxxxxxx
http://lists.squid-cache.org/listinfo/squid-users
