Hi,
Can you tell me the time it will take when I send a short message from one
node to another? Because we need the exact time when transfering buffer
between some nodes.
BTW,we use QNX 4.25.
Thanks,
Zhoujun
That’s totaly unrelated to ticksize (the subject).
Time is kind of unpredictable, there are too many variables,
it depends on network speed, network traffic, network equipment, etc.
“jgp” <gpjin@nairc.ac.cn> wrote in message news:9c2sbh$e8j$1@inn.qnx.com…
Hi,
Can you tell me the time it will take when I send a short message from one
node to another? Because we need the exact time when transfering buffer
between some nodes.
BTW,we use QNX 4.25.
Thanks,Zhoujun
Mario Charest <mcharest@deletezinformatic.com> wrote:
That’s totaly unrelated to ticksize (the subject).
Time is kind of unpredictable, there are too many variables,
it depends on network speed, network traffic, network equipment, etc.
What he said. The answer is anywhere from very short time to never.
Ethernet bandwidth (and, thereby throughput) is allocated in a
stochastic manner – there is no theoretical maximum latency.
Your best way to determine this, is to write a benchmark program that
does a lot of little Send/Receive/Reply transactions between the nodes
in question with the expected network load for your target environment
and benchmark the results. And, even this will only give you a
reasonable guess.
“jgp” <> gpjin@nairc.ac.cn> > wrote in message news:9c2sbh$e8j$> 1@inn.qnx.com> …
Hi,
Can you tell me the time it will take when I send a short message from one
node to another? Because we need the exact time when transfering buffer
between some nodes.
BTW,we use QNX 4.25.
Thanks,Zhoujun
\
-David
QNX Training Services
dagibbs@qnx.com
The tx time depends on a number of variables, but it can be made to be
predictable. If you have a point to point connection between two nodes
using ethernet, you should insure the driver is in full duplex mode.
With 100 Mbit ethernet it should be possible to characterize the
transmission time with an accuracy within the single digit millisecond
range (I would guess that 1 ms should be possible).
-----Original Message-----
From: jgp [mailto:gpjin@nairc.ac.cn]
Posted At: Monday, April 23, 2001 8:39 PM
Posted To: os
Conversation: ticksize
Subject: ticksize
Hi,
Can you tell me the time it will take when I send a short message from
one
node to another? Because we need the exact time when transfering buffer
between some nodes.
BTW,we use QNX 4.25.
Thanks,
Zhoujun
What he said. The answer is anywhere from very short time to never.
Ethernet bandwidth (and, thereby throughput) is allocated in a
stochastic manner – there is no theoretical maximum latency.
This is not true. Point to point full duplex ethernet connections do
not have any media access control at all, and are therefore
deterministic. Of course, as soon as there is some sort of media access
control (i.e. hub/switch) then it is indeterminate (since there are no
published standards for arbitration within a switch).
Your best way to determine this, is to write a benchmark program that
does a lot of little Send/Receive/Reply transactions between the nodes
in question with the expected network load for your target environment
and benchmark the results. And, even this will only give you a
reasonable guess.
I agree about the method, but if you control the environment
sufficiently you should be able to do much better than a reasonable
guess (you should be able to get an absolute worst case time - assuming
error free transmission). If there is an error in transmission, the
fact that you know the worst case tx time, allows you to have a
deterministic worst case time with single event error.
Rennie Allen <RAllen@csical.com> wrote:
What he said. The answer is anywhere from very short time to never.
Ethernet bandwidth (and, thereby throughput) is allocated in a
stochastic manner – there is no theoretical maximum latency.This is not true. Point to point full duplex ethernet connections do
not have any media access control at all, and are therefore
deterministic.
True, my statement is not valid for the special case you describe
above. In MOST ethernet configurations/installations, the allocation
is stochastic.
Of course, as soon as there is some sort of media access
control (i.e. hub/switch) then it is indeterminate (since there are no
published standards for arbitration within a switch).
Actually, you don’t even need a hub/switch – if you have several nodes
sharing a section of thinlan, you don’t have a hub/switch but you still
have collision behaviour.
-David
QNX Training Services
dagibbs@qnx.com
This is not true. Point to point full duplex ethernet connections do
not have any media access control at all, and are therefore
deterministic.True, my statement is not valid for the special case you describe
above. In MOST ethernet configurations/installations, the allocation
is stochastic.
Yes. I simply wanted to make clear that a deterministic ethernet
configuration can be built (which inherently shows that QNX is capable
of such a set up). Also point-to-point is not as useless as it sounds,
you can quite easily have 8 cards in a “master” box with 8 slave nodes
in a deterministic network. With low-end ethernet adapters this quite
cost effective for many distributed control applications (cell
automation comes to mind).
Of course, as soon as there is some sort of media access
control (i.e. hub/switch) then it is indeterminate (since there are
no
published standards for arbitration within a switch).Actually, you don’t even need a hub/switch – if you have several
nodes
sharing a section of thinlan, you don’t have a hub/switch but you
still
have collision behaviour.
Yup, although 10base2 is even less common than point to point 10/100
base T 