Web Service Performance

I developed a Web Application that has a Web Service component. I also
created a direct API to the second layer of the Web Service without going
through the Http connection to the Web Service. I noticed that if I make the
calls through the direct API (getting about 4KB of data from disk), it will
take between 100-200 ms. However, if I make the calls via the normal
[WebMethod] Http interface, the same calls takes between 500-700 ms. That
seems excessive since the url is localhost.

Since I am using localhost in the url, I would think that the performance
shouldn't be greatly affected by line speed or is the transfer speed limited
by the line speed even though it is internal? Any ideas why there would be
such a great difference in performance between the direct API call and the
http call?

ranjan
Sat Jan 29 03:36:29 CST 2005

On Fri, 28 Jan 2005 21:59:01 -0800, mterzich wrote:

> Since I am using localhost in the url, I would think that the performance
> shouldn't be greatly affected by line speed or is the transfer speed limited
> by the line speed even though it is internal? Any ideas why there would be
> such a great difference in performance between the direct API call and the
> http call?

Serialization + wire(normally not noticable for a localhost address) +
Deserialization
--


http://dotnetjunkies.com/weblog/dotnut

mterzich
Sat Jan 29 07:19:01 CST 2005

If your saying the wire shouldn't be much of an issue, then the
serialization/deserialization must be the poorest written code ever since it
adds between 1/4 and 1/2 second to the operation to transmit and receive a
4KB to 8KB data set. To me a poorly wtitten serialization/deserialization
operation shouldn't take more than 5 ms on a 3.0 giga hertz P4 machine. It
still sounds like it is the wire since I'm still in testing on my Windows
2000 Pro which is on a cable modem with 3.0Mb/128Kb external speed. At the
128Kb upload speed (about 12KB), that would explain the extra 1/4 to 1/2
second.

"ranjan.listserv@gmail.com" wrote:

> On Fri, 28 Jan 2005 21:59:01 -0800, mterzich wrote:
>
> > Since I am using localhost in the url, I would think that the performance
> > shouldn't be greatly affected by line speed or is the transfer speed limited
> > by the line speed even though it is internal? Any ideas why there would be
> > such a great difference in performance between the direct API call and the
> > http call?
>
> Serialization + wire(normally not noticable for a localhost address) +
> Deserialization
> --
>
>
> http://dotnetjunkies.com/weblog/dotnut
>

Frank
Sat Jan 29 09:29:25 CST 2005

We also found web service deserialization to be extremely slow in .net,
especially if you are transferring a lot of data. If you need to update
often, it is best to use TCP/IP if you can (sockets), or direct http if you
cannot.

Regards,
Frank Hileman

check out VG.net: http://www.vgdotnet.com
Animated vector graphics system
Integrated Visual Studio .NET graphics editor

"mterzich" <mterzich@discussions.microsoft.com> wrote in message
news:0AA869D8-6C70-463E-A31F-CFB7D55FBAB8@microsoft.com...
> If your saying the wire shouldn't be much of an issue, then the
> serialization/deserialization must be the poorest written code ever since
> it
> adds between 1/4 and 1/2 second to the operation to transmit and receive
> a
> 4KB to 8KB data set. To me a poorly wtitten serialization/deserialization
> operation shouldn't take more than 5 ms on a 3.0 giga hertz P4 machine. It
> still sounds like it is the wire since I'm still in testing on my Windows
> 2000 Pro which is on a cable modem with 3.0Mb/128Kb external speed. At the
> 128Kb upload speed (about 12KB), that would explain the extra 1/4 to 1/2
> second.
>
> "ranjan.listserv@gmail.com" wrote:
>
>> On Fri, 28 Jan 2005 21:59:01 -0800, mterzich wrote:
>>
>> > Since I am using localhost in the url, I would think that the
>> > performance
>> > shouldn't be greatly affected by line speed or is the transfer speed
>> > limited
>> > by the line speed even though it is internal? Any ideas why there would
>> > be
>> > such a great difference in performance between the direct API call and
>> > the
>> > http call?
>>
>> Serialization + wire(normally not noticable for a localhost address) +
>> Deserialization
>> --
>>
>>
>> http://dotnetjunkies.com/weblog/dotnut
>>

Steve
Sat Jan 29 13:04:36 CST 2005

serialization is quite fast, except runtime serialization on objects like
datasets.

When request is made, it goes through (i may be wrong with some of these
steps but you will get the picture) HTTP as a POST, IIS passes this to
ASP.NET, ASP.NET runs your CODE, your CODE returns data, the DATA is
serialized as sent to the requestor via HTTP the requestor desterilizes the
OBJECT.

if you want to test serialization on its own, you could do some tests,
also..... web service in general scale better, eg one request may seam slow
but when you start adding load its starts to perform very well compared to
other remote methods.

but..... you should not use SOAP on the same server? why would you do this?

Steve


"mterzich" <mterzich@discussions.microsoft.com> wrote in message
news:0AA869D8-6C70-463E-A31F-CFB7D55FBAB8@microsoft.com...
> If your saying the wire shouldn't be much of an issue, then the
> serialization/deserialization must be the poorest written code ever since
it
> adds between 1/4 and 1/2 second to the operation to transmit and receive
a
> 4KB to 8KB data set. To me a poorly wtitten serialization/deserialization
> operation shouldn't take more than 5 ms on a 3.0 giga hertz P4 machine. It
> still sounds like it is the wire since I'm still in testing on my Windows
> 2000 Pro which is on a cable modem with 3.0Mb/128Kb external speed. At the
> 128Kb upload speed (about 12KB), that would explain the extra 1/4 to 1/2
> second.
>
> "ranjan.listserv@gmail.com" wrote:
>
> > On Fri, 28 Jan 2005 21:59:01 -0800, mterzich wrote:
> >
> > > Since I am using localhost in the url, I would think that the
performance
> > > shouldn't be greatly affected by line speed or is the transfer speed
limited
> > > by the line speed even though it is internal? Any ideas why there
would be
> > > such a great difference in performance between the direct API call and
the
> > > http call?
> >
> > Serialization + wire(normally not noticable for a localhost address) +
> > Deserialization
> > --
> >
> >
> > http://dotnetjunkies.com/weblog/dotnut
> >

B
Sun Jan 30 21:08:56 CST 2005

Of course it's very dependant on the parameters you are passing. If these
are complex types (e.g. datasets) then serialisation will be slower.

Ben


"Steve Drake" <Steve@_NOSPAM_.Drakey.co.uk> wrote in message
news:u8j$jVjBFHA.4072@tk2msftngp13.phx.gbl...
> serialization is quite fast, except runtime serialization on objects like
> datasets.
>
> When request is made, it goes through (i may be wrong with some of these
> steps but you will get the picture) HTTP as a POST, IIS passes this to
> ASP.NET, ASP.NET runs your CODE, your CODE returns data, the DATA is
> serialized as sent to the requestor via HTTP the requestor desterilizes
the
> OBJECT.
>
> if you want to test serialization on its own, you could do some tests,
> also..... web service in general scale better, eg one request may seam
slow
> but when you start adding load its starts to perform very well compared to
> other remote methods.
>
> but..... you should not use SOAP on the same server? why would you do
this?
>
> Steve
>
>
> "mterzich" <mterzich@discussions.microsoft.com> wrote in message
> news:0AA869D8-6C70-463E-A31F-CFB7D55FBAB8@microsoft.com...
> > If your saying the wire shouldn't be much of an issue, then the
> > serialization/deserialization must be the poorest written code ever
since
> it
> > adds between 1/4 and 1/2 second to the operation to transmit and receive
> a
> > 4KB to 8KB data set. To me a poorly wtitten
serialization/deserialization
> > operation shouldn't take more than 5 ms on a 3.0 giga hertz P4 machine.
It
> > still sounds like it is the wire since I'm still in testing on my
Windows
> > 2000 Pro which is on a cable modem with 3.0Mb/128Kb external speed. At
the
> > 128Kb upload speed (about 12KB), that would explain the extra 1/4 to 1/2
> > second.
> >
> > "ranjan.listserv@gmail.com" wrote:
> >
> > > On Fri, 28 Jan 2005 21:59:01 -0800, mterzich wrote:
> > >
> > > > Since I am using localhost in the url, I would think that the
> performance
> > > > shouldn't be greatly affected by line speed or is the transfer speed
> limited
> > > > by the line speed even though it is internal? Any ideas why there
> would be
> > > > such a great difference in performance between the direct API call
and
> the
> > > > http call?
> > >
> > > Serialization + wire(normally not noticable for a localhost address) +
> > > Deserialization
> > > --
> > >
> > >
> > > http://dotnetjunkies.com/weblog/dotnut
> > >
>
>

William
Sun Jan 30 21:37:55 CST 2005

Also don't measure the first time as you don't want to include assembly load
time (xml, xmlserializer, etc) in your test or avg time. The next ones will
be much faster.

--
William Stacey, MVP
http://mvp.support.microsoft.com

"B S Wootton" <ben@nospam.com> wrote in message
news:#bJo2I0BFHA.2180@TK2MSFTNGP10.phx.gbl...
> Of course it's very dependant on the parameters you are passing. If these
> are complex types (e.g. datasets) then serialisation will be slower.
>
> Ben
>
>
> "Steve Drake" <Steve@_NOSPAM_.Drakey.co.uk> wrote in message
> news:u8j$jVjBFHA.4072@tk2msftngp13.phx.gbl...
> > serialization is quite fast, except runtime serialization on objects
like
> > datasets.
> >
> > When request is made, it goes through (i may be wrong with some of these
> > steps but you will get the picture) HTTP as a POST, IIS passes this to
> > ASP.NET, ASP.NET runs your CODE, your CODE returns data, the DATA is
> > serialized as sent to the requestor via HTTP the requestor desterilizes
> the
> > OBJECT.
> >
> > if you want to test serialization on its own, you could do some tests,
> > also..... web service in general scale better, eg one request may seam
> slow
> > but when you start adding load its starts to perform very well compared
to
> > other remote methods.
> >
> > but..... you should not use SOAP on the same server? why would you do
> this?
> >
> > Steve
> >
> >
> > "mterzich" <mterzich@discussions.microsoft.com> wrote in message
> > news:0AA869D8-6C70-463E-A31F-CFB7D55FBAB8@microsoft.com...
> > > If your saying the wire shouldn't be much of an issue, then the
> > > serialization/deserialization must be the poorest written code ever
> since
> > it
> > > adds between 1/4 and 1/2 second to the operation to transmit and
receive
> > a
> > > 4KB to 8KB data set. To me a poorly wtitten
> serialization/deserialization
> > > operation shouldn't take more than 5 ms on a 3.0 giga hertz P4
machine.
> It
> > > still sounds like it is the wire since I'm still in testing on my
> Windows
> > > 2000 Pro which is on a cable modem with 3.0Mb/128Kb external speed. At
> the
> > > 128Kb upload speed (about 12KB), that would explain the extra 1/4 to
1/2
> > > second.
> > >
> > > "ranjan.listserv@gmail.com" wrote:
> > >
> > > > On Fri, 28 Jan 2005 21:59:01 -0800, mterzich wrote:
> > > >
> > > > > Since I am using localhost in the url, I would think that the
> > performance
> > > > > shouldn't be greatly affected by line speed or is the transfer
speed
> > limited
> > > > > by the line speed even though it is internal? Any ideas why there
> > would be
> > > > > such a great difference in performance between the direct API call
> and
> > the
> > > > > http call?
> > > >
> > > > Serialization + wire(normally not noticable for a localhost address)
+
> > > > Deserialization
> > > > --
> > > >
> > > >
> > > > http://dotnetjunkies.com/weblog/dotnut
> > > >
> >
> >
>
>