[cisco-voip] SRST and standby SUB?

Justin Steinberg jsteinberg at gmail.com
Mon May 11 15:36:55 EDT 2009


ok, thanks guys for clearing that up.    Looking at a packet capture from an
idle phone all I see is the SCCP KA and the SCCP KA ACK every 30 seconds by
default so there is some dependency on the SCCP KA traffic generating the
required TCP traffic for Geometric TCP to work.

So theoritically, with your example of 2 ms RTT, if the CM server were to
lose power at the beginning of the 30 second KA cycle then the IP phone
would failover after approximately 30.030 seconds with Geometric TCP enabled
instead of 61.8 seconds with Geometric TCP disabled.

Sorry Scott, didn't mean to take your thread a different direction I just
found the Geometric TCP interesting.

Justin

On Mon, May 11, 2009 at 2:50 PM, Wes Sisk <wsisk at cisco.com> wrote:

>  you're catching on.  have another espresso and you'll be there.
>
> TCP is almost completely independent of the SCCP keepalives.
>
> this tends to help folks:
> SCCP keepalives verify the CM process is still active and processing
> traffic.  think "application layer"
> TCP works at the transport/session layer to verify data makes it over the
> network.
>
> this only becomes confusing because the 'network' doesn't really do
> anything by itself.  the 'application' must initiate data.  at that point
> the network must reliably deliver the data.  data can be signaling for an
> inbound call, signaling for an outbound call, signaling for an MWI update,
> signagling for a shared line update, or data for an SCCP keepalive.
>
> purely hypothetical numbers here:
> Geometric TCP:
> tcp retransmits very rapidly at multiples of round trip time.  If normal
> round trip time is 2msec the phone may retransmit at 2, 4, 8, and 16 msec.
> If phone does not receive a TCP ACK within 2+4+8+16msec = 30msec the phone
> gives up on the TCP session and fails over.
>
> "Slow Failover" or non-Geometric:
> TCP retransmits at 250,400,750, 1400, 2000, 4000,8000,15000 msec
> (CSCed01179).  Sum = 31.8 seconds.  Phone can wait 31 seconds for a TCP ACK
> from CM before giving up on TCP session and failing over.
>
> Geometric TCP is perceived particularly poorly if you have fast but
> unreliable network.
>
> /wes
>
>
> On Monday, May 11, 2009 2:24:46 PM, Justin Steinberg
> <jsteinberg at gmail.com> <jsteinberg at gmail.com> wrote:
>
> nevermind, what i wrote doesn't make any sense.
>
> I assume it must work something more like, the phones follow the SCCP
> keepalive timers defined in the ccmadmin service parameters but when the KA
> ACKs begin to fail the phone must use some 'geometric TCP' logic to force a
> quicker failover.
>
> On Mon, May 11, 2009 at 2:21 PM, Justin Steinberg <jsteinberg at gmail.com>wrote:
>
>> interesting.  I must have missed this new addition to 7.2(1).   So, this
>> means that with Geometric TCP enabled the phones don't really use the SCCP
>> keepalive timers configured in the CCMADMIN service parameters section?
>> They basically come up with a baseline TCP RTT and failover after that
>> interval passes three times without any response?
>>
>> On Mon, May 11, 2009 at 1:18 PM, Wes Sisk <wsisk at cisco.com> wrote:
>>
>>> One more variable I do not see in Ryan's response - Geometric TCP.  With
>>> Geometric TCP enabled the phones failover MUCH more aggressively.  With this
>>> phone may determine CM is down with only 1.5 second network outage.  With
>>> Geometric TCP disabled phone may wait up to 45 seconds to classify cm as
>>> "down" and attempt failover.
>>>
>>> CSCsm81227.
>>>
>>> /Wes
>>>
>>> On Monday, May 11, 2009 11:55:08 AM, Ryan Ratliff <rratliff at cisco.com>
>>> wrote:
>>>
>>>> So coming back from SRST there are 3 things that have to happen before
>>>> the phone will re-register.
>>>> 1. Establish TCP connection to one of the CMs
>>>> 2. Wait out the "Connection Monitor Duration" timer so it knows the
>>>> connection is stable (Enterprise Params, default 120 secs)
>>>> 3. Request and receive a registration token from the server indicating
>>>> that it can proceed with registration.
>>>>
>>>> Both connection monitor duration and the registration token mechanisms
>>>> are in place to make sure the failback process is as fast and problem-free
>>>> as possible.
>>>>
>>>> In your case is there anything between the sites that could be proxying
>>>> the tcp connection from phone to CCM?  All "Standby" means is that there's a
>>>> TCP session established but it's not actually registered.
>>>>
>>>> Your best bet is going to be to reset the phone and get a packet
>>>> capture.  look at the traffic on tcp 2000 to the CCM servers to see what's
>>>> going on.
>>>>
>>>> -Ryan
>>>>
>>>> On May 11, 2009, at 11:43 AM, Scott Voll wrote:
>>>>
>>>> no
>>>> sub
>>>> pub
>>>> srst
>>>>
>>>>
>>>> all via IP addresses
>>>>
>>>> sub -- Standby
>>>> pub
>>>> SRST -- Active
>>>>
>>>> Scott
>>>>
>>>> On Mon, May 11, 2009 at 8:41 AM, Ryan Ratliff <rratliff at cisco.com>
>>>> wrote:
>>>> On the phone what is the list of Communications Manages?  Does it
>>>> somehow think the SRST is higher in priority than the sub?
>>>>
>>>> -Ryan
>>>>
>>>>
>>>> On May 11, 2009, at 11:29 AM, Scott Voll wrote:
>>>>
>>>> I'm banging my head on a wall.  What am I missing?
>>>>
>>>> have a remote site over IPSEC VPN connection back to central CM cluster.
>>>>
>>>> all has been working fine. but today I find that the three phones (all)
>>>> are in SRST mode.  the VGW is still registered to the cluster. but from the
>>>> web interface of the phones shows reg'd with SRST and standby with SUB.  How
>>>> does that work?
>>>>
>>>> I don't really understand the logs on the 7942.
>>>>
>>>> I can ping from the sub and pub to the phones.  any ideas?
>>>>
>>>> I can attach logs from the phone is that helps.
>>>>
>>>> Thanks
>>>>
>>>> Scott
>>>> _______________________________________________
>>>> cisco-voip mailing list
>>>> cisco-voip at puck.nether.net
>>>> https://puck.nether.net/mailman/listinfo/cisco-voip
>>>>
>>>>
>>>>
>>>> _______________________________________________
>>>> cisco-voip mailing list
>>>> cisco-voip at puck.nether.net
>>>> https://puck.nether.net/mailman/listinfo/cisco-voip
>>>>
>>>
>>> _______________________________________________
>>> cisco-voip mailing list
>>> cisco-voip at puck.nether.net
>>> https://puck.nether.net/mailman/listinfo/cisco-voip
>>>
>>
>>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://puck.nether.net/pipermail/cisco-voip/attachments/20090511/8fc19056/attachment.html>


More information about the cisco-voip mailing list