Re: issue 381: Discovery of the support of the HTTP2 protocol: DNS-based Upgrade

On 28 February 2014 06:46, Amos Jeffries <squid3@treenet.co.nz> wrote:
> However, we are still facing complaints that for some URLs a whole
> timeout span (seconds/minutes) is waited for even though one of the
> results has been returned in a few ms. This has been directly tracked
> down to DNS resolvers simply not responding to one query out of the
> pair. Usually the IPv6 response from broken DNS resolvers.
>
> In practice the barrier waiting makes this algorithm *worse* in
> performance than the happy-eyeballs algorthim for the case where the
> timing difference of lookups matters most.

This seems like a key problem to me. UAs that are browser-like will be
able to mitigate the latency involved here by doing things like DNS
prefetching, but non-browser UAs and proxies don't have that luxury.
In particular, a mandate that conforming HTTP/2.0 clients MUST wait
for an SRV lookup to timeout before proceeding will kill clients like
mine stone-dead by providing a gargantuan potential performance hit.
This will mean that either we choose not to conform, or we give up and
get our lunch eaten by non-conforming implementations.

If we're determined to use SRV for discovery (and I see no reason not
to), happy-eyeballs seems like the better route to me.

Cory

Received on Friday, 28 February 2014 09:00:03 UTC