[zeromq-dev] Any update on Precision Time Protocol (PTP)?

Ilya Dmitrichenko errordeveloper at gmail.com
Wed Sep 28 13:21:33 CEST 2011


On 28 September 2011 07:23, Noospheer Team <team at noospheer.org> wrote:
> A couple weeks back there was talk of integrating PTP into 0MQ for A/V
> synchronization purposes. I imagine such an effective timing system
> would be also useful for making 0MQ HPC friendlier.

I have started that thread.

I am currently in the position of getting the hardware working, I have obtained
a board with Marvell Kirkwood chip, and the PTP block in the NIC core of this
SoC appears to be supported by current Linux kernel.

The only thing I am missing at the moment is the JTAG, which I have to use
due to the fact that I seem to have messed up the boot loader somehow.

You can contact me directly if you are interested to discuss this and, perhaps,
do some work together :)

As I mentioned, there exists software-only implementation (ptpv2d), though
I don't think it would make a huge difference and my intention is to use PTP
with hardware support.

Certainly, the patents may become an issue, though it doesn't stop anyone
from experimenting with it under GPL, and find a solution once it goes on
sale (well, it might be quite complicated though).

Although it is probably a good solution to integrate 0MQ and PTP from the
top down, however I presume that a closer coupling done with a PTP-aware
transport layer should give a greater advantage in theory.

As an example, if you look at how PTP is used in A/V briding (AVB), there
exists P802.1Qav ("Forwarding and Queuing Enhancements for Time-Sensitive
Streams") and essentially what it does is aligning of A/V packets within the
buffers of every switch in fully-compliant AVB network. However the experiments
have shown that end-points can achieve fairly reasonable results
without P802.1Qav.

Please have a look at the documentation of XMOS AVB reference design:
http://www.xmos.com/applications/avb
Just in case if you are interested in source code, it's is available here:
http://github.com/xcore/sw_avb

Another conclusion I'd like to make, there is probably not much use for PTP
on the Internet scale, let me know if you see anything that disapproves that.
Perhaps those who are interested (including myself) should join here:
http://lists.infradead.org/mailman/listinfo/ptpd
or rather here: http://groups.google.com/group/ptpv2d/subscribe?note=1

Cheers,
-- 
Ilya



More information about the zeromq-dev mailing list