?
=?ISO-8859-1?Q?J=F6rn_Petersen?=
Hi everybody!
Currently we are deploying our devices with NT embedded, but we plan
to upgrade to XP. We have created an XPe image meeting our special
demands and done some tests, where we compare the UDP performance of
the NTe and the XPe platforms.
The test is a very simple UDP-server (running on the embedded target)
that binds a socket to a specific port, and then in a loop receives
messages and responds to the client as fast as possible.
The client program (running on my desktop machine) is executing a loop
that sends datagrams to the server, waits for the response and counts
the number of replies in a certain period.
The result of this test (same server application running on the same
HW platform sitting in the same switch - the only difference is the
OS) is as follows:
Target running XPe replies about 136.000 times in a minute
Target running NTe replies about 197.000 times in a minute
We have made the same test on two desktop machines running XP Pro and
NT and they end up with very close results, so nothing is wrong with
XP in general.
We have done other tests with a program called QCheck (from NetIQ -
see www.netiq.com/qcheck for details about this program). This tool
measures the throghput between two endpoints and using UDP the
throughput to the NTe target is usually on 37 Mbps (sometimes dropping
to 34 Mbps), while the XPe target varies a lot between 22 and 35 Mbps.
The tool also measures the response time of the endpoint and while NTe
always responds in 7 ms on a 32.000 bytes packet, XPe usually averages
to 9 ms with an observed maximum latency of 21 ms.
So our conclusion is, that we have made some mistake in our XPe design
and setup or that we need to configure XPe in a certain way to achieve
the performance of NTe (which is a requirement for us to upgrade the
OS).
And this is where I need your help, since I haven't got a single clue
about how to trace this down. Does anybody have an idea what to do or
where to search?
Currently we are deploying our devices with NT embedded, but we plan
to upgrade to XP. We have created an XPe image meeting our special
demands and done some tests, where we compare the UDP performance of
the NTe and the XPe platforms.
The test is a very simple UDP-server (running on the embedded target)
that binds a socket to a specific port, and then in a loop receives
messages and responds to the client as fast as possible.
The client program (running on my desktop machine) is executing a loop
that sends datagrams to the server, waits for the response and counts
the number of replies in a certain period.
The result of this test (same server application running on the same
HW platform sitting in the same switch - the only difference is the
OS) is as follows:
Target running XPe replies about 136.000 times in a minute
Target running NTe replies about 197.000 times in a minute
We have made the same test on two desktop machines running XP Pro and
NT and they end up with very close results, so nothing is wrong with
XP in general.
We have done other tests with a program called QCheck (from NetIQ -
see www.netiq.com/qcheck for details about this program). This tool
measures the throghput between two endpoints and using UDP the
throughput to the NTe target is usually on 37 Mbps (sometimes dropping
to 34 Mbps), while the XPe target varies a lot between 22 and 35 Mbps.
The tool also measures the response time of the endpoint and while NTe
always responds in 7 ms on a 32.000 bytes packet, XPe usually averages
to 9 ms with an observed maximum latency of 21 ms.
So our conclusion is, that we have made some mistake in our XPe design
and setup or that we need to configure XPe in a certain way to achieve
the performance of NTe (which is a requirement for us to upgrade the
OS).
And this is where I need your help, since I haven't got a single clue
about how to trace this down. Does anybody have an idea what to do or
where to search?