[cisco-voip] Transcoding/Conferencing on the same ISDN Gateway

Sun Feb 14 12:20:33 EST 2010

I should probably note that the new ASR platform has a new packet
processing architecture and this doesn't quite apply to that box.  I'm
guessing it's also why it can handle 5000 calls or so.

-nick

On Sun, Feb 14, 2010 at 12:19 PM, Nick Matthews <matthnick at gmail.com> wrote:
> Yes, it's true that all packets to CUBE are processed switched.  I was
> just making the distinction between passing a certain amount of PPS,
> because simply processing the PPS is normally the published number for
> processors.  When you add an MTP into the call flow, and whether or
> not the CUBE is the MTP also, it doubles the amount of RTP packets to
> be processed by the CPU (nearly).  There is some efficiency involved,
> but it still comes out to 35-45%, rather than 50.
>
> The numbers I'm talking about are calls.  The CUBE performance numbers
> are done in calls or sessions, not legs.  Twice as many legs, and
> actually some may call them sessions as well.  2 legs = 1 call.  We're
> talking 400 calls.
>
> -nick
>
> On Fri, Feb 12, 2010 at 1:40 AM, Rhodium <rhodium_uk at yahoo.co.uk> wrote:
>> Thanks for replying. You are either up early or up late... :)
>>
>>> What really kills the CPU is the PPS that go
>>> directly to the
>>> CPU because the RTP packets are targeted to a directly
>>> connected IP
>>> address.
>>
>> Isn't that true for a normal call in flow-through mode because it has to terminate and re-initiate calls from a local interface and hence everything will end up process switched anyway. So essentially, for arguement's sake lets say that a CUBE supports 400 simultaneous sessions disregarding the CPS for now, and we have 200 sessions for MTP configured, then it would stand to reason that realistically, we can only support a maximum of 200 channels and not even that as there is the processing overhead of SCCP as the 200 sessions of MTP would count as logical RTP channels.
>>
>> Also something came to mind, if a particular CUBE or ISDN GW supports 400 sessions, then in the case of the CUBE would 400 sessions be for incoming call legs or 200 incoming, and 200 outgoing and the same fo an ISDN GW?
>>
>> Regards,
>>
>> J
>>
>> Not quite sure I understand the difference between a call that terminates on a CUBE for example,
>>
>> --- On Fri, 2/12/10, Nick Matthews <matthnick at gmail.com> wrote:
>>
>>> From: Nick Matthews <matthnick at gmail.com>
>>> Subject: Re: [cisco-voip] Transcoding/Conferencing on the same ISDN Gateway
>>> To: "Rhodium" <rhodium_uk at yahoo.co.uk>
>>> Cc: "Matthew Saskin" <msaskin at gmail.com>, cisco-voip at puck.nether.net
>>> Date: Friday, February 12, 2010, 6:28 AM
>>> These are from general numbers I've
>>> looked at in terms of CUBE
>>> scalability.  It's not as drastic as 50%, and by that
>>> I mean more like
>>> 35-40%.  What really kills the CPU is the PPS that go
>>> directly to the
>>> CPU because the RTP packets are targeted to a directly
>>> connected IP
>>> address. It's the exact same reason why you get 10-15% more
>>> usage out
>>> of vad on, because there is a linearly scaled number fewer
>>> packets
>>> with that feature on.
>>>
>>> -nick
>>>
>>> On Thu, Feb 11, 2010 at 11:13 PM, Rhodium <rhodium_uk at yahoo.co.uk>
>>> wrote:
>>> > Hi Nick,
>>> >
>>> > Thanks for the info. It re-iterates what is stated in
>>> the SRND and what I know of CPS.
>>> >
>>> > Quite an interesting fact about how MTP, etc
>>> approximately halve the number of calls, I assume
>>> simultaneous calls? Didn't think it was that drastic. Any
>>> documentation on that or just "experience"?
>>> >
>>> > Regards,
>>> >
>>> > J
>>> >
>>> > --- On Fri, 2/12/10, Nick Matthews <matthnick at gmail.com>
>>> wrote:
>>> >
>>> >> From: Nick Matthews <matthnick at gmail.com>
>>> >> Subject: Re: [cisco-voip] Transcoding/Conferencing
>>> on the same ISDN Gateway
>>> >> To: "Rhodium" <rhodium_uk at yahoo.co.uk>
>>> >> Cc: "Matthew Saskin" <msaskin at gmail.com>,
>>> cisco-voip at puck.nether.net
>>> >> Date: Friday, February 12, 2010, 1:42 AM
>>> >> FWIW, if you're looking at 5+ CPS
>>> >> you're probably looking at an
>>> >> AS5X000 box.  Many of the concurrent call numbers
>>> that
>>> >> are calculated
>>> >> for gateways are not done at that high of a call
>>> rate
>>> >> (implying short
>>> >> duration calls).
>>> >>
>>> >> Regardless, you can run everything you want on the
>>> router
>>> >> (VPN, MTP,
>>> >> Xcoding, full BGP table) as long as the parameters
>>> of the
>>> >> equation are
>>> >> right.  If you scale your gateway the only
>>> problem is
>>> >> CPU usage, not
>>> >> feature overload..  The only addendum to that is
>>> that
>>> >> most of these
>>> >> numbers are calculated individually, so when you
>>> mix and
>>> >> match the
>>> >> calculations aren't quite as accurate.  Which
>>> doesn't
>>> >> say much,
>>> >> because depending on your call patterns you may
>>> get
>>> >> different results
>>> >> from the posted numbers anyways.  Long calls =
>>> better
>>> >> for CPU, vad on
>>> >> = better for CPU, codec selection = doesn't
>>> matter.
>>> >> Most of the
>>> >> problem you run into is the pure PPS of these
>>> streams
>>> >> rather than the
>>> >> overhead, which is why MTPs and such can get you
>>> higher CPU
>>> >> pretty
>>> >> quickly.  Xcoding and MTP roughly halve the
>>> amount of
>>> >> calls you can
>>> >> make through the box.
>>> >>
>>> >> hth,
>>> >> nick
>>> >>
>>> >> On Wed, Feb 10, 2010 at 5:21 PM, Rhodium <rhodium_uk at yahoo.co.uk>
>>> >> wrote:
>>> >> > http://www.cisco.com/en/US/docs/voice_ip_comm/cucm/srnd/6x/gateways.html#wp1043594
>>> >> >
>>> >> > I am referring to CPS and not simultaneous
>>> call
>>> >> volumes. We have multiple gateways decked with 210
>>> channels
>>> >> each (7 x E1) for redundancy and have higher call
>>> volumes
>>> >> but we are focused on CPS as that brings down the
>>> number of
>>> >> simultaneous calls on the gateways.
>>> >> >
>>> >> > Also our BHCA is in the region of 8000 so
>>> there is a
>>> >> slight difference in type of traffic flow we are
>>> seeing.
>>> >> >
>>> >> > Regards,
>>> >> >
>>> >> > J
>>> >> >
>>> >> > --- On Wed, 2/10/10, Matthew Saskin <msaskin at gmail.com>
>>> >> wrote:
>>> >> >
>>> >> >> From: Matthew Saskin <msaskin at gmail.com>
>>> >> >> Subject: Re: [cisco-voip]
>>> Transcoding/Conferencing
>>> >> on the same ISDN Gateway
>>> >> >> To: "Rhodium" <rhodium_uk at yahoo.co.uk>
>>> >> >> Cc: "Peter Slow" <peter.slow at gmail.com>,
>>> >> cisco-voip at puck.nether.net
>>> >> >> Date: Wednesday, February 10, 2010, 10:01
>>> PM
>>> >> >> Where in the SRND is the call limit
>>> >> >> for various gateways listed?  I'd like
>>> to take a
>>> >> >> look.
>>> >> >>
>>> >> >> The environment I was referring to was a
>>> UCCE
>>> >> build
>>> >> >> performed by Cisco AS.  4 gateways w/
>>> 10-12 PRI's
>>> >> each
>>> >> >> handling upwards of 50K calls/day inbound
>>> and
>>> >> outbound - all
>>> >> >> calls were transcoded on these gateways
>>> after the
>>> >> IVR and
>>> >> >> before heading to agents.  During busy
>>> hour call
>>> >> volumes
>>> >> >> weren't above 250 calls/gateway but I'd
>>> like to
>>> >> see
>>> >> >> the reference.
>>> >> >>
>>> >> >>
>>> >> >> Matthew Saskin
>>> >> >> msaskin at gmail.com
>>> >> >> 203-253-9571
>>> >> >>
>>> >> >> July 18, 2010 - 1500m swim (in the
>>> hudson), 40k
>>> >> bike, 10k
>>> >> >> run
>>> >> >> Please support the Leukemia & Lyphoma
>>> Society
>>> >> >>
>>> >> >> http://pages.teamintraining.org/nyc/nyctri10/msaskin
>>> >> >>
>>> >> >>
>>> >> >>
>>> >> >> On Wed, Feb 10, 2010 at 4:53 PM,
>>> >> >> Rhodium <rhodium_uk at yahoo.co.uk>
>>> >> >> wrote:
>>> >> >>
>>> >> >> Hi Matthew,
>>> >> >>
>>> >> >> Appreciate the feedback. I am just
>>> concerned that
>>> >> as this
>>> >> >> is for a call centre, when the number of
>>> calls per
>>> >> second
>>> >> >> (CPS) increases, the number of
>>> simultaneous calls
>>> >> >> decreases... so I can see a deployment
>>> with 1 to 3
>>> >> CPS
>>> >> >> working with the setup you have described
>>> but with
>>> >> something
>>> >> >> close to around 15 CPS, then according to
>>> the
>>> >> SRND, that is
>>> >> >> a maximum of 255 calls (for a 3845). As
>>> we are
>>> >> decking out
>>> >> >> the ISDN router with 210 channels, then
>>> that only
>>> >> leaves
>>> >> >> headroom for 45 calls not including CPU
>>> overhead
>>> >> of SCCP,
>>> >> >> which led to my mistaken recollection of
>>> >> transcoders and
>>> >> >> ISDN GWs when it was transcoders and
>>> CUBEs.
>>> >> >>
>>> >> >>
>>> >> >> I guess based on the setup I described,
>>> the more I
>>> >> think
>>> >> >> about it, the more I am inclined to make
>>> it "off
>>> >> >> box". Am I over riding financial
>>> judgement with
>>> >> caution
>>> >> >> too much?
>>> >> >>
>>> >> >> Regards,
>>> >> >>
>>> >> >> J
>>> >> >>
>>> >> >>
>>> >> >>
>>> >> >> --- On Wed, 2/10/10, Matthew Saskin
>>> <msaskin at gmail.com>
>>> >> >> wrote:
>>> >> >>
>>> >> >> > From: Matthew Saskin <msaskin at gmail.com>
>>> >> >>
>>> >> >> > Subject: Re: [cisco-voip]
>>> >> >> Transcoding/Conferencing on the same ISDN
>>> Gateway
>>> >> >> > To: "Rhodium" <rhodium_uk at yahoo.co.uk>
>>> >> >> > Cc: "Peter Slow" <peter.slow at gmail.com>,
>>> >> >> cisco-voip at puck.nether.net
>>> >> >>
>>> >> >> > Date: Wednesday, February 10, 2010,
>>> 9:37 PM
>>> >> >>
>>> >> >>
>>> >> >>
>>> >> >> > I regularly see single gateways
>>> >> >> handle
>>> >> >> > all functions (PSTN termination,
>>> >> conferencing,
>>> >> >> transcoding)
>>> >> >> > for sites of all sizes.  I know of
>>> nothing
>>> >> >> [design-wise]
>>> >> >> > preventing you from doing this.  In
>>> fact, I
>>> >> can think
>>> >> >> of
>>> >> >>
>>> >> >> > some very large installations I've
>>> seen that
>>> >> were
>>> >> >> done
>>> >> >> > by advanced services where 3845's
>>> were being
>>> >> used
>>> >> >> to
>>> >> >> > terminate 10-12 PRI's plus handle a
>>> few
>>> >> hundred
>>> >> >> > transcoding sessions.
>>> >> >>
>>> >> >> >
>>> >> >> >
>>> >> >> > Matthew Saskin
>>> >> >> > msaskin at gmail.com
>>> >> >> > 203-253-9571
>>> >> >> >
>>> >> >> > July 18, 2010 - 1500m swim (in the
>>> hudson),
>>> >> 40k bike,
>>> >> >> 10k
>>> >> >> > run
>>> >> >> > Please support the Leukemia &
>>> Lyphoma
>>> >> Society
>>> >> >>
>>> >> >> >
>>> >> >> > http://pages.teamintraining.org/nyc/nyctri10/msaskin
>>> >> >> >
>>> >> >> >
>>> >> >> >
>>> >> >> > On Wed, Feb 10, 2010 at 4:24 PM,
>>> >> >> > Rhodium <rhodium_uk at yahoo.co.uk>
>>> >> >>
>>> >> >> > wrote:
>>> >> >> >
>>> >> >> > Hi Peter,
>>> >> >> >
>>> >> >> > Thank you for your reply.
>>> >> >> >
>>> >> >> > I dug out a report from Cisco
>>> Advanced
>>> >> Services that
>>> >> >> stated
>>> >> >> > that CUBEs should only be used for
>>> SIP trunks
>>> >> and not
>>> >> >> for
>>> >> >>
>>> >> >> > transcoding/mtp. Knew I read
>>> something
>>> >> somewhere. Just
>>> >> >> not
>>> >> >> > the details... :)
>>> >> >> >
>>> >> >> >
>>> >> >> > So we are all definite that for an
>>> ISDN GW,
>>> >> we can
>>> >> >> use
>>> >> >> > re-use the DSPs for transcoding in
>>> line with
>>> >> >> Cisco's
>>> >> >>
>>> >> >> > recommendations? I want to recommend
>>> getting
>>> >> a
>>> >> >> dedicated
>>> >> >> > router for these functions but need
>>> to
>>> >> justify the
>>> >> >> cost so
>>> >> >> > if there are no design
>>> recommendations, then
>>> >> I guess I
>>> >> >> can
>>> >> >> > just put it all on one box.
>>> >> >>
>>> >> >> >
>>> >> >> >
>>> >> >> > Regards,
>>> >> >> >
>>> >> >> > J
>>> >> >> >
>>> >> >> >
>>> >> >> > --- On Wed, 2/10/10, Peter Slow
>>> <peter.slow at gmail.com>
>>> >> >> > wrote:
>>> >> >> >
>>> >> >> > > From: Peter Slow <peter.slow at gmail.com>
>>> >> >>
>>> >> >> >
>>> >> >> > > Subject: Re: [cisco-voip]
>>> >> >> Transcoding/Conferencing on
>>> >> >> > the same ISDN Gateway
>>> >> >> > > To: "Rhodium" <rhodium_uk at yahoo.co.uk>
>>> >> >> > > Cc: cisco-voip at puck.nether.net
>>> >> >>
>>> >> >> >
>>> >> >> > > Date: Wednesday, February 10,
>>> 2010, 5:02
>>> >> PM
>>> >> >> >
>>> >> >> >
>>> >> >> >
>>> >> >> > > That is not necessarily
>>> correct. It
>>> >> >> > > depends heavily on the type of
>>> >> >> > > gateway you're talking about,
>>> and the
>>> >> number
>>> >> >> of
>>> >> >>
>>> >> >> > transcoding
>>> >> >> > > or
>>> >> >> > > conferencing sessions you might
>>> be
>>> >> needing.
>>> >> >> >
>>> >> >> > >
>>> >> >> > > -peter
>>> >> >> > >
>>> >> >> > > On Wed, Feb 10, 2010 at 7:24
>>> AM, Rhodium
>>> >> <rhodium_uk at yahoo.co.uk>
>>> >> >>
>>> >> >> > > wrote:
>>> >> >> > > > Hi Experts, :)
>>> >> >> > > >
>>> >> >> > > > I am sure I read somewhere
>>> in a
>>> >> design doc
>>> >> >> or the
>>> >> >> > SRND
>>> >> >> >
>>> >> >> > > that it is not advocated to
>>> put
>>> >> transcoding or
>>> >> >> > conferencing
>>> >> >>
>>> >> >> > > resources on a voice gateway
>>> handling
>>> >> about 150
>>> >> >> > calls.
>>> >> >> > > >
>>> >> >> > > > Am I recalling right or
>>> are the old
>>> >> brain
>>> >> >> cells
>>> >> >> > > getting weaker with age.
>>> >> >> >
>>> >> >> > > >
>>> >> >>
>>> >> >> > > > If that is correct, a link
>>> would
>>> >> be
>>> >> >> appreciated
>>> >> >> > as I
>>> >> >> > > can't find it.
>>> >> >> > > >
>>> >> >> > > > Regards,
>>> >> >> > > >
>>> >> >> > > > Jason
>>> >> >> > > >
>>> >> >> > > >
>>> >> >>
>>> >> >> > > >
>>> >> >> > > >
>>> >> >> >
>>> >> >> > > >
>>> >> >>
>>> _______________________________________________
>>> >> >> > > > 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
>>> >> >
>>> >>
>>> >
>>> >
>>> >
>>> >
>>> >
>>>
>>
>>
>>
>>
>>
>