[erlang-questions] spdy in mouchi web

James Churchman jameschurchman@REDACTED
Sat May 7 15:21:51 CEST 2011


>Some of the concepts remind me of an experiment I made called VCC (
http://svn.ulf.wiger.net/vcc/trunk/vcc), which >also implemented "streams"
on top of VCC (although I called them "virtual channels", corrupted by years
of work in >ATM networking design). I incorporated Erlang notions like
monitoring channels, and also "spawning" temporary >channels based on a
"template", and close-on-reply (which SPDY also has a version of).

as usual erlang / erlangers are years ahead with ideas over the general
status quo :-)

>Distributed Erlang has no means of prioritizing things like net_kernel ping
messages...

not the kindof aplication i was thinking of but you woudl know more than me
Ulf :-)

>It'll become interesting whenever the protocol becomes stable and
>whenever they allow Chrome to use it outside the Google services,

>it's apparently hardcoded for everything else currently.

i did not know this but it sounds sensible... if it was widely implemented
then proved to have some fatal problem, google can work this out by testing
it its self on quite a good use case like gmail, then fixing it ( websockets
anyone ? )

im sure if the experiment is successful, even if there was a minor
modification then google would open it up to all

> ( it doesn't give much advantage over pipelining )

are you sure it does not give much? and for the kind of web-apps built in
erlang might benifit most!

> so its adoption in other browsers might take a while.

browser makers are totaly obsessed with sub microsecond benchmarks ( even if
they don't effect the end user much along with css selectors that no-one
ever uses ) so i think it would be implemented in firefox very fast... maybe
not IE, but most users on IE are not as discriminating users!



On 7 May 2011 13:13, Loïc Hoguin <essen@REDACTED> wrote:

> On 05/07/2011 12:09 PM, James Churchman wrote:
> > So my question is ( might be quid a huge ask), is, does anybody plan to
> > add this to mouciweb ( and therefore webmachine )
> >
> > I am guessing its a bit of a chicken/egg seniario but if it has google
> > behind it that may help adoption, and it may bring dev's to erlang if
> > its (one of) the first to support it!
>
> It'll become interesting whenever the protocol becomes stable and
> whenever they allow Chrome to use it outside the Google services, as
> it's apparently hardcoded for everything else currently.
>
> It is an interesting experiment as it allows processing multiple
> requests in parallel over a single TCP connection. However, for most
> applications it doesn't give much advantage over pipelining so its
> adoption in other browsers might take a while.
>
> Still this is interesting enough to support it whenever we can actually
> use it, in my opinion.
>
> --
> Loďc Hoguin
> Dev:Extend
>


On 7 May 2011 12:17, Ulf Wiger <ulf.wiger@REDACTED> wrote:

>
> Interesting.
>
> As SPDY sits on top of SSL, one might also note that Erlang's (new) SSL
> implementation is unusually scalable - not everyone who wants to run a web
> server will be happy to put an SSL offload accelerator in front of it…
>
> But reading further, the current version of SPDY uses TCP instead.
>
> Some of the concepts remind me of an experiment I made called VCC (
> http://svn.ulf.wiger.net/vcc/trunk/vcc), which also implemented "streams"
> on top of VCC (although I called them "virtual channels", corrupted by years
> of work in ATM networking design). I incorporated Erlang notions like
> monitoring channels, and also "spawning" temporary channels based on a
> "template", and close-on-reply (which SPDY also has a version of).
>
> But the main reason for my experiment was to support version negotiation
> between peer nodes, together with side-effect-free version transformers, in
> order to handle redundancy upgrades in large Erlang systems.
>
> Perhaps the SPDY ideas of uni-directional streams and re-prioritization of
> streams could hold some promise also for erlang-to-erlang communication?
> Currently, Distributed Erlang has no means of prioritizing things like
> net_kernel ping messages...
>
> BR,
> Ulf W
>
> On 7 May 2011, at 12:09, James Churchman wrote:
>
> > So my question is ( might be quid a huge ask), is, does anybody plan to
> add this to mouciweb ( and therefore webmachine )
> >
>
> Ulf Wiger, CTO, Erlang Solutions, Ltd.
> http://erlang-solutions.com
>
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://erlang.org/pipermail/erlang-questions/attachments/20110507/5f800343/attachment.htm>


More information about the erlang-questions mailing list