thrift-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Ben Craig <bencr...@apache.org>
Subject Re: TPipe and TPipeServer for C#
Date Mon, 17 Mar 2014 15:34:30 GMT
Your mileage may vary :)

Those tests were run on a fairly low end machine, that was freshly 
reimaged.  That machine tends to recompile all the .NET assemblies in the 
background while the tests run.  If you send large messages, you will 
probably get different performance characteristics as well.

One of the things that I really like about named pipes is that I don't 
have to deal with the limited resource of port numbers.  No "service 
location", IANA, or picking a number out of a hat and hoping no-one else 
uses it.  Just pick a sufficiently long name (maybe even a string-ified 
GUID), and you're good.

Joseph Fradley <joe.fradley@fradeng.com> wrote on 03/17/2014 10:26:07 AM:

> From: Joseph Fradley <joe.fradley@fradeng.com>
> To: user@thrift.apache.org, 
> Date: 03/17/2014 10:26 AM
> Subject: Re: TPipe and TPipeServer for C#
> 
> Thank you for running the comparison. So, we're looking at about ~20% 
speed
> improvement, not super awesome but still noticeable. I think in the 
short
> term I'll move forward with sockets. If (when) I hit a performance issue
> I'll look at the named pipe option.
> 
> 
> On Mon, Mar 17, 2014 at 11:20 AM, Ben Craig <bencraig@apache.org> wrote:
> 
> > For an idea of the performance difference...
> > With C++, using sockets, I can do 10000 "trivial" IPC calls in 
0.328708
> > seconds.
> > With C++, using named pipes, I can do 10000 "trivial" IPC calls in
> > 0.263701 seconds.
> >
> > Both using framed transports, and the threaded server.
> >
> > Joseph Fradley <joe.fradley@fradeng.com> wrote on 03/17/2014 10:08:28 
AM:
> >
> > > From: Joseph Fradley <joe.fradley@fradeng.com>
> > > To: user@thrift.apache.org,
> > > Date: 03/17/2014 10:08 AM
> > > Subject: Re: TPipe and TPipeServer for C#
> > >
> > > Great, thanks. I'll create a JIRA feature ticket for it and I'll 
give it
> > a
> > > whirl over the next few days.
> > >
> > > Having named pipe transport between a C# app and a C++ app should 
give
> > > better performance then sockets (when on the same machine of 
course).
> > >
> > >
> > >
> > > On Mon, Mar 17, 2014 at 11:02 AM, Carl Yeksigian 
<carl@yeksigian.com>
> > wrote:
> > >
> > > > Hi Joe,
> > > >
> > > > I had started to work on this over a year ago; I've rebased and 
pushed
> > it
> > > > up here:
> > > >
> > > > https://github.com/carlyeks/thrift/commit/
> > > 313e0e2d3d46607ce1185ce55e3d1c4ebce03241
> > > > .
> > > > I never needed to use it, so I never finished or tested it. There
> > isn't a
> > > > JIRA, so feel free to create one.
> > > >
> > > > -Carl
> > > >
> > > >
> > > > On Mon, Mar 17, 2014 at 9:57 AM, Joseph Fradley
> > <joe.fradley@fradeng.com
> > > > >wrote:
> > > >
> > > > > Hello,
> > > > >
> > > > > Are there current plans to bring the Named Pipe transport over 
to
> > C#?
> > > > > Currently, it appears to only be available for C++. I searched 
the
> > > > feature
> > > > > tickets and didn't find anything.
> > > > >
> > > > > Thank you,
> > > > > Joe
> > > > >
> > > >
> >
> >


Mime
View raw message