[Beowulf] Question on clock skew of LANai

[Dan Kidger]

Dear Song,
   I suspect that your best way to get answers to these in-depth question is 
to ask Myricom directly. (Unless of course you are planning on competing with 
them.:-) )

If you had a signal analyser then perhaps you copuld calculate some of these 
values yourself.  .. and remember most high performacne interconnect do *not* 
do store and forward of packets as it adds too much to the latency as well as 
requiring significant local memory.

Also have you read their whitepapers on their website?   

Yours,
Daniel.

--------------------------------------------------------------
Dr. Dan Kidger, Quadrics Ltd.      daniel.kidger at quadrics.com
One Bridewell St., Bristol, BS1 2AA, UK         0117 915 5505
----------------------- www.quadrics.com --------------------


On Thursday 22 July 2004 9:51 am, 송효정 wrote:
> Hi.
>
> I am Hyo Jung Song from Samsung Electronics.
> For our research & development,
> we are very interested in Myrinet switches.
>
> At this time,
> we need some information on Myrinet-2000.
>
> Specifically, we want to know following information.
> If you give me some details,
> it would be helpful to proceed our R&D.
>
>
> For Small Myrinet-2000 Switches, I would like to know the *actual value* of
> the below parameter :
>
>
> - routing delay
> the time for the Myrinet-2000 Switch to process the header flit of a packet
> (i.e. to decide the output link depending on the routing info. in the
> header flit) plus the time for the switch (crossbar) to transfer the header
> flit from the input link to the assigned output link
>
>
> - switching delay
> ; the time for the switch to transfer a non-header(data) flit from the
> input link to the assigned output link
>
>
>
> - flow control delay
> ; Assuming that stop and go flow control is used for the link level flow
> control.
>
> a. the time for the sender switch to generate the stop flit upon some
> predefined condition (i.e. buffer occupancy) (the time excludes the actual
> tranfer time to transfer the stop flit over the link, and counts delay only
> inside the switch)
>
> b. the time for the receiver switch to decide not to send any more data
> upon the stop flow control.
>
>
> I suppose there is a flow controller inside a switch to implement stop and
> go control. Basically, what i want to know is *the time to operate flow
> controller inside a switch* upon "stop" part. And I assume the time is
> almost same as go part.
>
>
>
>
>
> Besides, as for the NIC, can we get the clock skew #s in the ppm unit?
>
>
> Thank you very much.
>
>
>
>
>
> Hyo Jung Song
> Senior Engineer
> Samsung Electronics
> tel. 82-2-3416-0355
> _______________________________________________
> Beowulf mailing list, Beowulf at beowulf.org
> To change your subscription (digest mode or unsubscribe) visit
> http://www.beowulf.org/mailman/listinfo/beowulf

--