Does PXI 6682 deal with asymmetric delays?

[geppetto]

Hi all

I am considering the adoption of NI PXI/PCI cards for clock synchronization

in a campus-range distributed system (a few miles extent).

I read in the best practices section of "NI PXI-6682 User Manual" (Oct. 2007), page 4-4:

"Use short cabling when possible. Ethernet cabling is inherently

asymmetric; the longer the cabling, the higher the asymmetry. This
impacts synchronization performance, because the IEEE 1588
protocol assumes a symmetric network path."

Indeed IEEE1588 is designed ti effectively cope with asymmetric delays,

so why 6682 doesn't.

Is there any experience about this in this Forum?

Thanks in advance!

PS: ah, also: any hope to see a Linux driver/software.

[JoshuaP]

Why do you think the spec is designed to cope with asymmetric delays?  In section 6.1.3 Assumptions on page12 of the IEEE 1588-2002 spec, which the PXI-6682 complies with it states "Network delay between master and slave on a subnet must be symmetric (see 7.8.1.2)" in order to achieve optimal clock synchronization performance.  The current data set for any clock only uses a single member for the one_way_delay, so we don't effectively have the ability to cope with asymmetric delays.

-Josh

[geppetto]

Hi  Joshua, and thank you for your prompt reply: I surely need some

guide for my unsure steps 🙂

You are right, it is not in protocol specifications (but unfortunately I was convinced of that).

I read about asymmetry compensation in more than one place. Here is one:

André Vallat, Dominik Schneuwly, "Clock Synchronization in Telecommunications via PTP (IEEE 1588)",

http://ieeexplore.ieee.org/iel5/4318993/4318994/04319093.pdf

I can send you the paper in case you cannot reach it. They describe a way to estimate asymmetry.

But I do not know whether it is of practical interest or not.

Given that asymmetry seems to be the main source of inaccuracy (and jitter when variable)

my guess is that it would be appropriate to guarantee a high precedence to PTP traffic and use

as much fiber than possible: do you agree?

Could you please tell me whether you know of a variant of PXI-6682/PCI-1588 with fiber

output (as far as I see there is only copper out).

Thanks for your patience...

[JoshuaP]

Thanks for the link.  I think the source of all of the confusion is the adoption of version 2 of the 1588 specification.  The PXI-6682 currently supports the IEEE 1588-2002 specification, but the paper your reference references the IEEE 1588-2008 specification which is commonly refered to as v2 by the industry.  The paper is right that the v2 specification has more in place to deal with assymetric delays.  Unfortunately, there aren't any fiber variants of the PXI/PCI-1588 boards, but I'm sure there are several copper to fiber adapters out there if that is what you are interested in using.

In general any network component that has a fifo/buffer will introduce jitter and the larger the buffer the more jitter the system will have based on network load changes.  From the testing that I have done the simplest network components like hubs have the best 1588 performance.  The dumber the hub the better the performance.  Another thing you can do is use a boundary clock or transparent switch, which is designed to composate for jitter through the switch.

http://zone.ni.com/devzone/cda/tut/p/id/2822

I don't want to speculate on any future products or enhancements, but if NI ever updates the PXI-6682 or PCI-1588 to support IEEE 1588-2008 it should be able to accomidate more asymmetric delays based on what I have seen in the v2 spec.

-Josh

[geppetto]

Yes, probably the best solution is to use converters near NI boards and fiber passive stars.

IMHO boundary clocks are not appropriate for applications that require high clock accuracy, since they

deteriorate synchronization (jitter and inaccuracy tend to accumulate each time you synchronize a clock).

On the other side transparent switches seem to be a niche product (I know of just one producer), but

with little plus with respect to a solution based on converters and fiber.

I'd be glad to know your opinion on this... Do you know of any experiment using fiber media?

Thank you