Extremely low buffer usage on perf monitor

GNU Radio
1

I am looking into CPU and buffer usage of my OOT module via CtrlPort
Performance Monitor.

I have two flow graphs, a transmitter and a receiver.

I can see a quite reasonable performance measures on the receiver side:

However, on the transmitter side, buffer usage shows very weird values:

It says, the transmitter of my OOT module rarely uses buffer. But, i
don’t
think so, and actually it uses buffers! At least, I think encoders on
the
transmitter should use buffers as much as decoders on the receiver does.
(Of course, it’s not true that encoder and decoder require same
computational cost. But, what I mean is that such tiny value is
ridiculous.)

I’ve executed the transmitter flow graph several times again and again,
that number (1.52…e-5) never changes.

My guess is, that number is a floating point representation of 1/(2^16).
And 2^16 is 64k. But I have no idea what it means.

Can anyone give me a tiny hint of it?

Regards
Jeon.

2

I’m sorry for something that I didn’t write on the previous post.

Comparing runtime and buffer, runtime and buffer have different number
of
blocks.
In many cases of mine, buffer shows a fewer blocks than runtime.

My guess is that, it’s because buffer of sink blocks don’t have to be
monitored.
Is what I guess is correct?

If so, this might be a problem when I want to monitor a certain block
which has an optional output signature so that it is recognized as a
sink
block.

Is there a way to monitor a such block?

Regards,
Jeon.

2015-08-06 19:31 GMT+09:00 Jeon [email protected]:

3

Hi Jeon,

only output buffers are counted – because they are the input buffers of
the next block “downstream” in your flow graph. Also, a single buffer
can only have one block writing to it, but multiple blocks reading from
it, so it’s logical to only count the “outputs”; hence, sinks don’t
count at all.
Would that explain the difference in numbers?

Best regards,
Marcus

4

On Thu, Aug 6, 2015 at 8:14 AM, Jeon [email protected] wrote:

If so, this might be a problem when I want to monitor a certain block
which has an optional output signature so that it is recognized as a sink
block.

Is there a way to monitor a such block?

Regards,
Jeon.

If it’s a sink block, then no, it doesn’t have an output buffer and so
cannot be monitored this way. If it has an output buffer, it will show
up
here.

I’ve executed the transmitter flow graph several times again and again,

Those are strange numbers. They should represent a percentage, so all of
the numbers in these tables should add up to 1. It’s possible that one
buffer is empty during some calls to get the performance counter and
we’re
not protecting against that, which is causing strange behavior (not
exactly
a divide-by-zero, but maybe similar?). So no, I don’t quite know.

Also, please be careful when sending images in your emails. They make
them
overly large, and our system might catch and block them instead of
letting
the pass. Better to either copy the values into a table or post this as
a
screenshot somewhere and send the link instead.

Tom

5

Just FYI, Jeon’s images are a link (to imgur.com).

Ron