[ntp:questions] OT: GPS18x LVC failure

Rick Jones rick.jones2 at hp.com
Mon Oct 12 21:19:07 UTC 2009


Unruh <unruh-spam at physics.ubc.ca> wrote:
> "David J Taylor" <david-taylor at blueyonder.not-this-part.nor-this.co.uk.invalid> writes:

> >"Unruh" <unruh-spam at physics.ubc.ca> wrote in message 
> >news:TtqAm.47651$Db2.44785 at edtnps83...
> >[]

> >> USB only has one data line. It cannot "do" a DCD line except as
> >> another bit of serial data, interleaved with all the other serial
> >> data.  Ie, you are not going to get any good signals for an
> >> "interrupt".

> >> You definitely need either a real serial port with a true
> >> hardware line, or a parallel port to get true interrupts.

> >Be that as it may, a GPS/PPS system working over USB provided
> >better results (45 microseconds jitter) on my test PC than a 1Gb/s
> >Ethernet card (110-140 microseconds jitter) on a 1Gb/s lightly
> >loaded LAN talking to a stratum-1 server, so I would not say that
> >USB is useless.

> Actually those Gb ethernet switches seems to have some really bad
> and variable latencies. My system, before they installed Gb
> switches, had a consistant roundtrip of 140 usec and gave better
> than 20usec jitter (with 100Mb switches). Since those new switches
> were installed things have gone to hell-- variable roundtrips (
> 140usec, 300usec and up to 1 sec roundtrip) and much much worse
> jitter. NOt sure where the problem is, but it sure is terrible.
> Doing a plot of roundtrip vs offset it is definitely one way delay.

I cannot speak to what happens in the switches, but in the NICs, I
think that almost all GbE NICs have some form of interrupt
coalescing/avoidance - the idea is to reduce the CPU utilization to
allow higher throughput.  This does nasty thing to latency, and could
be somewhat variable depending on how much other traffic is flowing
through the NIC.  In linux-land, some variations on ethtool commands
and/or modprobe parameters would be involved in diminishing if not
outright disabling the interrupt coalescing.

ftp://ftp.cup.hp.com/dist/networking/briefs/nic_latency_vs_tput.txt

is an old writeup, but it should give the gist of it.

As for things switches, when and what sort of spanning tree is done,
and perhaps lifetimes for the MAC forwarding tables come to mind.

rick jones
-- 
oxymoron n, Hummer H2 with California Save Our Coasts and Oceans plates
these opinions are mine, all mine; HP might not want them anyway... :)
feel free to post, OR email to rick.jones2 in hp.com but NOT BOTH...




More information about the questions mailing list