- From: James M Snell <jasnell@gmail.com>
- Date: Wed, 10 Jun 2009 09:25:13 -0700
- To: Patrick McManus <mcmanus@ducksong.com>
- CC: Daniel Stenberg <daniel@haxx.se>, Bjoern Hoehrmann <derhoermi@gmx.net>, HTTP Working Group <ietf-http-wg@w3.org>
Patrick McManus wrote: > "Once the batched request messages have been processed, a single HTTP > Batch Response message is created wherein each part contains the HTTP > Response Messages." > > This seems to mean the server cannot begin to generate any response > until all portions of the request have been received. That would seem to > recreate the same head of line blocking problem on a connection we > already have with pipelines. > > I know this is probably not the answer you'd like to hear but that's actually not a problem we're looking to solve with this. > "To simplify client processing of the response, the ordering of > response messages in the Batch Response SHOULD match the ordering of > request messages in the Batch Request." > > huh? The value I see in your proposal is that it allows out of order > responses which could improve latency for requests that would other wise > be "stuck-in-line". So why would you recommend against your strongest > (only?) feature. The client has to correlate the ID's anyhow - so its > not like you've made it easier. > > The next version of the draft will remove this SHOULD and allow the responses to be in any order. - James
Received on Wednesday, 10 June 2009 16:25:39 UTC