All, At 2016-03-10 17:15:12 +0800 Davey Song <songlinj...@gmail.com> wrote:
> FYI. A simple lab test done by my colleague. > > http://www.dnsv6lab.net/2016/03/05/A-performance-test-of-DNS-over-different-transport-protocol/ > > There are some observations： > 2) When coming to HTTPS, the keep-alive cannot reduce latency > significantly. The reason is probably that TLS has to do the handshake to > change the key which might be the main cost of latency. It is an unfortunate thing is that TLS adds so much time. My guess is that because of virtual hosting support in HTTP the TLS session is re-negotiated for each query (using SNI), requiring additional RTT. I wonder if it might be possible for a library to cache the previously used domain name and re-use a TLS session if it is the same? This could provide a win in the general case, probably with a simple LRU expiration to kick out TLS sessions and migrate to new ones if a new domain name is being used. This seems like the kind of thing that HTTP/2 folks would have already thought of, but it doesn't look like the Go library implements this sort of thing, just based on the performance measurements. Cheers, -- Shane
pgpbcuMoX94sa.pgp
Description: OpenPGP digital signature
_______________________________________________ DNSOP mailing list DNSOP@ietf.org https://www.ietf.org/mailman/listinfo/dnsop
