Re: Very rapid polling
- From: mayer@xxxxxxx (Danny Mayer)
- Date: Mon, 16 Feb 2009 04:23:04 GMT
Uwe Klein wrote:
Richard B. Gilbert wrote:
Uwe Klein wrote:
gary.limanapier@xxxxxxxxxxxx wrote:
On Feb 15, 9:15 am, Uwe Klein <uwe_klein_habertw...@xxxxxxxxxxx>
wrote:
gary.limanap...@xxxxxxxxxxxx wrote:
Just joining this discussion as we have noticed NTP request from our
Authorised Server every 300ms on our LAN. They occur only between the
Time Server and the WiFi Tablet during a socket connection while there
is a low signal strength. The Tablet sends its own NTP request with a
Poll Interval of 4 (16 seconds). The server returns its response
increasing the Poll Interval to 10 (1024 secs). The server then keeps
sending this same packet every 300-500ms.
Does the ntp process submit these packets repeatedly or
are you seeing some (uncalled for) "collision/resend" action due to
issues with the wlan connection?
uwe
Would it matter if the packet colided? NTP is using UDP and I can't
see from the packet analyser that it is expects a response back. When
it does work with a known good machine there is no acknowledgement of
the packet being sent. Do you think that there another level of
operation in the network that I am not seeing with the packet analyser?
you usually get retransmission on collision from the hardware ( or the
driver ).
( ethernet i am certain about, no firm knowledge about wlan though )
This would be independent of the protocol used.
this might help ( look for collision .. RTS/CTS packet ):
http://wlan.nat.sdu.dk/802_11standard.htm
From my side this is an absolutely blind guess. But it could
well be a point to look at before trying to find an app bug
that may not be there.
uwe
Remember that NTP uses UDP. UDP does not guarantee delivery, does not
resend packets, etc. It sends a response to a request and doesn't know
or care if the response was received or not!
Watch my lips ( or this wrongfull usage of that idiom) :
The hardware does automatic resend on collision using an
exponential+random backoff algorithm.
In Ethernet hardware this is buried in the chipset.
Wlan does similar stuff but buried in the driver OS side and/or Firmware side.
( And the newfangled stuff tends to have more and nefarious implementation issues )
It has nothing to do with UDP, TCP or whatever type of protocol you think you
are using ( wrong layer ).
An eternity ( in internet time ) ago I used to design ethernet interface cards.
You could have very subtle errors in the transceiver or cabling where some party
on the cable saw a collision while others did not.
Ethernet and friends do backoff/retry etc. but that doesn't mean that
the packet gets received by its destination. Lots of things can happen
in between including hardware failures, misrouting, firewalls, etc.
However at the IP protocol level UDP does not bother to resend if it
doesn't get an acknowledgement of receipt of the packet. TCP does care
but we are only using UDP.
Danny
uwe.
- References:
- Very rapid polling
- From: jlevine
- Re: Very rapid polling
- From: Richard B. Gilbert
- Re: Very rapid polling
- From: Unruh
- Re: Very rapid polling
- From: Richard B. Gilbert
- Re: Very rapid polling
- From: Unruh
- Re: Very rapid polling
- From: Richard B. Gilbert
- Re: Very rapid polling
- From: Unruh
- Re: Very rapid polling
- From: gary . limanapier
- Re: Very rapid polling
- From: Uwe Klein
- Re: Very rapid polling
- From: gary . limanapier
- Very rapid polling
- Prev by Date: Re: NTP over redundant peer links, undetected loops
- Next by Date: Re: NTP over redundant peer links, undetected loops
- Previous by thread: Re: Very rapid polling
- Next by thread: Re: Very rapid polling
- Index(es):
Relevant Pages
|
