22 Nov
2004
22 Nov
'04
5:59 p.m.
Thanks for the reply. See my inline comments.
Regards,
Paul
--- "Greger V. Teigre" <greger(a)teigre.com> wrote:
Thank you for sharing your views on using ser for
carrier-grade services. I
find that open source projects often lack information/sharing on
scalability, redundancy, monitoring and other important things in a
carrier-grade solution. Such information is often considered a competitive
advantage with the result that the open source project looses a lot of the
potential carrier-grade improvements.
I share with you the belief that ser can be put into production for
carrier-grade services. My company has a focus on end-user centric service
delivery and provide managed services, as well as turn-key solutions for
service providers. Our focus so far has been on home/mobile office services
with dial-up, VPN, (two-factor) authentication etc. Our platform has a
powerful and flexible service authentication and authorization engine at the
core, with RADIUS as one front-end (using LDAP backend). In addition, we
have a web front-end for end-user self-service and service administration,
as well as an XML interface for CRM integration. Finally, we have a
Windows-based software that can automatically configure a user's PC, install
components, keep them updated, and also be a "menu" or guide for the user in
using the services (voicemail, forwarding etc).
Our service providers often have multiple agents, channels, resellers.
Each with branding requirements, as well as CRM system integration
requirements. So, several virtual providers can be enabled on the same
platform. Of course, providing billing records is a must...
We are currently adding VoIP as a service option in our platform. I am
sure many people on this list are working on commercial projects each with a
slightly different angle. :-)
So more directly to the perspectives in your email:
- Interesting that you have selected WhiteBox as the distro. Surely, a
RedHat subscription at $349 should not really add a lot to the costs? We
have looked at both CentOs, another RHEL-based distro, and WhiteBox, but
find that the $349 is pretty cheap for the "luxury" of not having to worry
about distros dying or introducing problems in libraries etc.
It's funny you mention this because just this morning we were discussing a router
update that RH
released that is not in WhiteBox yet. So now we're considering going to to production
with RHPW
which is only $69 and comes with a free upgrade to REL 4 when available. RHPW is a bit of
a
backdoor loophole in the RedHat pricing structure.
- You have opted for the development version of ser.
We have been
(conservatively) staying with the 0.8.14. Any particular reason for going
for development? (Except for cool features; I think the AVP functionality is
particularly interesting) What is the experienced stability? Will you put
the development version into carrier production?!
Putting -dev17 into production is a possiblity. This really depends on when ser-0.9 stable
gets
released. If we can hold back the business until then we'll use stable 0.9 otherwise
we're forced
to use -dev17 (Which has been very stable in our testing). The reason for this is that we
have
dependencies on avpops and a few other modules that are not in 0.8.14
- SEMS is a bit immature and I have honestly been in
doubt whether I should
spend any time on it at this stage. We have looked at Asterisk as an option
(no time for religious wars even though there is some overlapping
functionality). In fact, we also evaluate PSTN-based voicemail solutions
for scalability and reliability. I must admit, it's a bit backwards, but...
Voicemail is a big problem right now. I know I'll make a few Asterisk advocates upset
with my next
comment - but our perception is our reality :-(. Asterisk is not ready for a
carrier-grade
implementation. Everyone we've seen and talked to that uses Asterisk has stability
issues when you
exceed a small to medium size company. We need to provide voicemail for as many as 500,000
users
and Asterisk ain't going to do the trick.
I'm not suggesting that sems/sipums will work either, but I believe it has a better
chance than
Asterisk.
If sems/sipums ends up not working either then we will be force to purchase carrier-grade
voicemail applicances that are SIP aware. :-(
- I'm surprised that you have set up serweb at
all. Our experience in
delivering end-user services is that your Achilles heel is support. There
is no limit to what a naive end-user can assume or believe and it will most
definitely cause you a support call.... I look at serweb as an example
application, and I think it would be very interesting to make a LGPL'ed
library for ser functions that could be accessible through soap services.
This way one can easily integrate ser functionality into the gazillion of
CRM and customer front-end systems that you find in the telco world.
serweb is a total band-aide until we create similar functionality in our J2EE web farm.
We've
stripped out much of what is in serweb and only give users access to things like their
personal
phonebook, speed dialing, etc.
- You mention an enum directory for subscribers.
I'm pretty sure we're going
there, but there quite some issues related to service provider roaming, how
it is charged etc. Of course, existing telcos are keen to stop any all-IP,
non-billable efforts, and those of us who deliver to telcos must (at least
to a certain degree) adhere to their demands. Non-incumbents will lead the
way here, but I believe that we need to establish a viable business model
for exchanging traffic. The market (or the providers) are not ready for
all-you-can-all-around-the-world options yet. As long as minutes are
charged between IP and PSTN, there will be an incentive to move as much
traffic as possible through PSTN (also for the non-incumbents with such
business models).
I agree.
Well, talking about rambling... I think the point is: By showing our cards
a bit, do we have some common interests that can benefit the ser community
or create a new business relationship?
Possibly. Voicemail is still my single biggest problem. And the only solutions I know of
are
Asterisk and sems. Do you know of any others (open source)?
As for helping the ser community - I really think the single biggest way would be to
product a
real book - kind of like a FSF GDB or GCC book with detailed explainations of "if
ser.cfg option
does this" and "if you need to provide xxxx then you must have yyyy in
ser.cfg" and expert hints
and things to avoid. I'm talking about a real book here not an abbrievated cookbook
or
mini-howto's.
Also, a paid consultancy like the ones that Orion Server or JBoss have would be awesome.
I've read
in the serusers list of people offering to pay someone to fix their ser.cfg.
That all said - I'm not an author and I know I'm not an authoritative
knowledgebase on ser so I'm
sure I'm not the one to do this, although I'd love to help. The sad part is that I
can only
imagine how busy Daniel, Jan, Jiri, Andrei and company are so I'd say nobody should
expect much
involvement from them, other than editing and probably most ser users aren't saavy
enough to read
ser source code to find out exactly how the product works. And (documentation == success)
of any
product since lack of documentation will lead to fewer deployments.
Just my US$0.02 worth
g-)
Java Rockx wrote:
=== message truncated ===
__________________________________
Do you Yahoo!?
Meet the all-new My Yahoo! - Try it today!
http://my.yahoo.com
Greger,
I believe the small change to the onfailure_route[] was the thing
that made the difference. Andrei pointed out a small error to me
where I was calling fix_nated_contact() for more messages than I
should have. (THANKS ANDREI!!!)
The other thing that made a **substantial** improvement was the fact
that our PSTN gateway provider that uses Sonus equipment as added as
an "alias" at the top of the ser.cfg. This caused ser to not include
"Route:" headers on ACKs and this was causing the Sonus equipment to
drop calls after two minutes.
It's funny how such small changes to ser.cfg can have dramatic
effects.
I'd love to see a complete book published on ser. I'm sure, with the
help of the maintainers and serusers participants this could become a
reality. I mean this is the best damned SIP router I know of.
Commercial solutions such as Sonus are not really a "solution"
because they run as much as US$2 million per box and most small
companies can't afford "one". And even if you could afford =>one<=
you may not be able to afford others for scaling your VoIP platform -
so whats the point of attempting VoIP if you can't scale?
Our solution is modeled after the Google network whereby:
unlimited horizontal scalablity == unlimited cps (calls per second)
And this equates to:
successful VoIP Company == retired employee
If we can indeed do this for our implementation and do it on low end
x86 or Opteron boxes (or blade servers), then the sky is the limit.
Our engineer whom we acquired from RedHat a few months ago already
has a full TCP/IP and UDP socket patch working so we can seperate ser
and serweb and scale them independently of each other (over a VPN)
since we don't need ser's existing FIFO.
SEMS is a whole other beast. We are now working on this using the ivr
module with sipums for a carrier-grade voicemail solution. Early
indications is that we still have plenty to do in order to get
99.999% reliablity with sems. Our goal is to not have sems run on the
same box as ser as we already know this is bad. Voicemail and SIP
proxy cannot be the same box if you expect to be a big VoIP provider,
IMHO.
So our high-level VoIP platform uses the following as independent
server farms:
MySQL 4.1.7 (farm 1)
ser-0.8.99-dev17 (farm 2)
serweb on Apache2 farm 3)
sems/ivr/sipums (farm 4, still a work in progress)
Third party PSTN gateway *** see comment below
All servers use Whitebox Linux Respin 1 and our network
implementation allows us to completely administer all servers at any
remote co-lo site --- even down to remote installation of the
operating system. This includes make a change to the master ser.cfg
and having it replicate to all ser proxies.
Monitoring is a big part of this network architecture and we've only
got basic monitoring right now, so this will become a focus of our
efforts soon.
We also use third parties for providing PSTN access. IMHO, PSTN
gateways are much more trouble than their worth. We'd rather rely on
multiple third parties for this than to try and do it ourselves. The
cost of DSP boards is expensive, especially if you wanted to handle
say 50,000 concurrent callers. For this very reason we use companies
such as Level 3 and Global Crossing and all we have to do is simply
monitor their QoS and route PSTN callers to other PSTN gateways as
needed. And this is all at the network level, so ser is totally
oblivious to this.
Another thing I've love to see is a database whereby VoIP service
providers can lookup 10-digit DIDs in a common database (much like a
root DNS server) and determine if a callee is a VoIP user and route
the SIP call directly from our SIP proxy to their SIP proxy rather
than making a SIP->PSTN->SIP call. We have the Chairman of the IPCC
deeply invovled in our company and he could get the ball rolling on
this, but we don't want to get distracted right now.
Anyhow, I've rambled long enough.
Regards,
Paul Hazlett
We'll publish our
--- "Greger V. Teigre" <greger(a)teigre.com> wrote:
> Paul,
> Thanks for posting your resulting config after so many hours of
> testing! I did a diff and can see that you have done several
> (smaller) changes. But what did the trick? (i.e. for Route problem)
> g-)
>
> Java Rockx wrote:
>> Greger,
>>
>> I finally got everything working without STUN and without an
>> Outbound Proxy. I an only using rtpproxy/nathelper.
>>
>> I've tested these settings with the following SIP Phones/ATAs
>>
>> Sipura
>> UTstarcom iAN-02EX
>> Grandstream ATA486
>> Grandstream BT100
>> Cisco ATA186
>> Cisco 7960G
>> WorldAccxx ATA
>> X-Ten Pro
>> X-Ten Lite
>>
>> We are very happy with everything now. The only piece of the puzzle
>> that I don't have working yet is sems/sipums for voicemail - but I'm
>> working on that.
>>
>> Attached is my complete ser.cfg file that is working. Please note
>> that I'm using ser-0.8.99-dev17 for this rather than ser-0.8.14
>>
>> If anyone finds problems in this ser.cfg then please let me know -
>> but like I said before, things seem very stable right now with
>> -dev17.
>>