Aside from the silliness of "can I ping you?", someone did make the valuable observation that ping might not be the best way to measure content delivery performance. A better way would be to measure the actual delivery performance by noting the size and delivery time for an actual bonafide request. The traffic directing system could default to random source servers (perhaps with coarse regional constraints) until enough data is gathered to guide future server decisions. As more client requests are received, occasional testing with other source servers could be done to ensure best server selection. A significant drop in throughput could also trigger some re-testing. An additional benefit would be a reasonable guess as to the bandwidth of the end user (dialup, broadband, etc). I see that Apache already has an optional custom-log attribute for transfer time in integer seconds (%T). It would be trivial to modify that to give milliseconds. If you could get this to work right it would be completely non-invasive and should produce better overall results. KL
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
Aside from the silliness of "can I ping you?", someone did make the
valuable observation that ping might not be the best way to measure content delivery performance.
I have heard some pretty bizarre proposals from certain vendors about what they've been thinking of doing.
A better way would be to measure the actual delivery performance by noting the size and delivery time for an actual bonafide request.
Hey, why bother measuring and all that stuff? TCP SYN comes in to load balancer, load balancer simply dispatches copies of the packet to all available real servers (with some fixups and spoofing the client's IP address), whichever one finishes the TCP handshake with the client first, is by definition, the fastest one. Hoo boy, I almost fell out of my chair when I heard this one described for the first time. I do not know if it was ever deployed in a shipping product. - --- "The avalanche has already begun. It is too late for the pebbles to vote" - Kosh -----BEGIN PGP SIGNATURE----- Version: PGPfreeware 6.5.8 for non-commercial use <http://www.pgp.com> iQA/AwUBO9nyG0ksS4VV8BvHEQLOSQCcCLOu3lqBoDsOz+MqUoCP7rIbROoAoKYj sjbr+ufFs3KBrxjujfNP5NEV =9AXe -----END PGP SIGNATURE-----
participants (2)
-
Kevin Loch -
Mike Batchelor