.NET Remoting Performance/Throughput

[Dave Griffin]

We're developing an client/server application where the client exports well
known services using remoting (using the TCP default formatter) and the
clients (usually there is only 1) attaches to these remoted classes and
dequeues data from a threadsafe queue. Functionally it works great, but the
amount of actual throughput we can get seems CPU limited. That is sending
large amounts of data such as serializable classes containing 5000 - 40000
element arrays of ushorts seems to bring both the client and the server to
their knees. Our experiments using the TCP/IP classes are a bit more
promising and seem to show that TCP/IP has a lot more thoughput. Is Remoting
really that inefficient in it's data transmission, and if so, is there
anything I can do about it (or do I have to use TCP/IP directly)?

Thanks!

[Champika Nirosh]

To broader your view .. Refer to these to articles
this..
http://www.codeproject.com/cs/webser...motingdime.asp
this .....
http://msdn.microsoft.com/library/de...tnetarch16.asp
and this
http://msdn.microsoft.com/library/de...tnetarch14.asp

Nirosh.

"Dave Griffin" <Dave Gr*****@discussions.microsoft.com> wrote in message
news:72**********************************@microsof t.com...

We're developing an client/server application where the client exports well known services using remoting (using the TCP default formatter) and the
clients (usually there is only 1) attaches to these remoted classes and
dequeues data from a threadsafe queue. Functionally it works great, but the amount of actual throughput we can get seems CPU limited. That is sending
large amounts of data such as serializable classes containing 5000 - 40000
element arrays of ushorts seems to bring both the client and the server to
their knees. Our experiments using the TCP/IP classes are a bit more
promising and seem to show that TCP/IP has a lot more thoughput. Is Remoting really that inefficient in it's data transmission, and if so, is there
anything I can do about it (or do I have to use TCP/IP directly)?

Thanks!

[Dave Griffin]

I've read the articles and I found them helpful. It makes me think that maybe
it's the default serialization of my classes that may be to blame. From
reading the articles I wonder if classes with my own serialization methods
might work better. I'm also not sure if using structures instead would be
better since I'm not clear about whether using classes results in references
being sent across rather than actual data.

"Champika Nirosh" wrote:

To broader your view .. Refer to these to articles
this..
http://www.codeproject.com/cs/webser...motingdime.asp
this .....
http://msdn.microsoft.com/library/de...tnetarch16.asp
and this
http://msdn.microsoft.com/library/de...tnetarch14.asp

Nirosh.

"Dave Griffin" <Dave Gr*****@discussions.microsoft.com> wrote in message
news:72**********************************@microsof t.com...

We're developing an client/server application where the client exports

well

known services using remoting (using the TCP default formatter) and the
clients (usually there is only 1) attaches to these remoted classes and
dequeues data from a threadsafe queue. Functionally it works great, but

the

amount of actual throughput we can get seems CPU limited. That is sending
large amounts of data such as serializable classes containing 5000 - 40000
element arrays of ushorts seems to bring both the client and the server to
their knees. Our experiments using the TCP/IP classes are a bit more
promising and seem to show that TCP/IP has a lot more thoughput. Is

Remoting

really that inefficient in it's data transmission, and if so, is there
anything I can do about it (or do I have to use TCP/IP directly)?

Thanks!