Quote:
|
Originally Posted by rdhw
The load-spreading within the cluster is done by destination URL, not by user. So if you always use the same test site for identifying your proxy, you will always get the same answer. Each member of the cluster has its own little bit if the internet to cache, and your connection will utilise all members of the cluster according to the URL sought. It's more efficient that way.
|
Absolutely! if the users of the same site were spread across all the caches, every cache would have to fetch a copy.
As for DNS timeout, I've seen my firewall catch attempts to issue ICMP destination unreachable, when the reply finally arrives too late to be any use.
If there's a way to tune this behaviour (in Win98SE), I've not found it yet.
What beats me, is that the address of NTL's email server (I might try plugging the ip address in directly) seems to be a victim as often as any remote address.
One thing which can disrupt DNS, is excessive load from firewall reverse DNS lookups (my firewall is set NOT to R-DNS every damn ping!, but many are).
That can magnify the normal DNS load by a huge margin.