TCP optimization

[Diego Woitasen]

On Fri, Feb 3, 2012 at 9:35 AM, Roy Warner <thisroywarner at gmail.com> wrote:
> Hi,
>
> We've been testing WANProxy for a while.
> I happen to ran into this today - http://en.wikipedia.org/wiki/SPDY
>
> And thought of this thread, and if this could help in any way to improve the
> speed.
> I'm not to technical, but just thought it's worth to share.
>
> All the Best,
> Roy
>
> On Fri, Feb 3, 2012 at 2:29 AM, Diego Woitasen <diego at woitasen.com.ar>
> wrote:
>>
>> On Tue, Jan 31, 2012 at 3:22 PM, Juli Mallett <juli at clockworksquid.com>
>> wrote:
>> > Hi Diego,
>> >
>> > A connection pool is indeed good and desirable, but I'm not sure about
>> > the overhead — I think we would need just as many roundtrips to
>> > establish the new session on an existing connection as to establish a
>> > new one; the thing which really saves is to put multiple proxied
>> > streams onto a single connection, but as you likely know, that breaks
>> > TCP's semantics somewhat — but it's still worth doing in some
>> > scenarios.
>>
>> Why do you think that there is performance problem? I think that to
>> maintain a pool of connections is easy and cheap (again, I think).
>> I'll try to experiment around this.
>>
>> Multiplexing is great, but it's more complex. You need to implement a
>> protocol inside TCP for stream identification, stream flow control,
>> etc.
>>
>> >
>> > Overall, I think multiplexing, connection pooling, etc., should all be
>> > easy to implement well and prove very useful in many circumstances.
>> >
>> > Thanks,
>> > Juli.
>> >
>> > On Tue, Jan 31, 2012 at 10:11, Diego Woitasen <diego at woitasen.com.ar>
>> > wrote:
>> >> Hi,
>> >>  In one of the customers where I'm going to implement Wanproxy we are
>> >> discussing about developing on-disk storage and tcp optimization.
>> >> On-disk storage was already discussed on this list, now I was to start
>> >> a discussion about TCP optimization.
>> >>
>> >>  Wanproxy optimizes the amount of data transferred right now, but one
>> >> of the biggest problems of some WAN connections is the latency
>> >> (Satellite specially). I was thinking about a method to improve TCP
>> >> using pools of connections.  Right now the Wanproxy client connects to
>> >> the Wanproxy server when it receives a TCP connection. I think that we
>> >> could improve the performance having a pool of idle TCP connections.
>> >> When a connection is established to the Wanproxy client, it finds a
>> >> free connection and use it. Then the connection could be closed and a
>> >> new one could be established to keep always the same numbers idle
>> >> ones. Or may be we could not close it and to be reused.
>> >>
>> >> In the past I worked in a consultancy job where I developed a proof of
>> >> concept of the described above and there results were great.
>> >>
>> >> What do you think?
>> >>
>> >> Regards,
>> >>  Diego
>> >>
>> >> --
>> >> Diego Woitasen
>> >> _______________________________________________
>> >> wanproxy mailing list
>> >> wanproxy at lists.wanproxy.org
>> >> http://lists.wanproxy.org/listinfo.cgi/wanproxy-wanproxy.org
>>
>>
>>
>> --
>> Diego Woitasen
>> _______________________________________________
>> wanproxy mailing list
>> wanproxy at lists.wanproxy.org
>> http://lists.wanproxy.org/listinfo.cgi/wanproxy-wanproxy.org
>
>

Looks interesting. That's why I posted to the list, to exchange ideas :)

Look like HTTP oriented, but may be it can be used for other protocols.

In the past, I've studied SCTP which is great (multiple streams/flows
per connection) but doesn't support flow control per stream. SPDY, in
a quick look, supports that. I'll read the specs carefully.

Thanks for the information.

-- 
Diego Woitasen