VoIP Systems

Telecom Exec: VOIP Is a Pain

CHICAGO – Supercomm 2005 – A Time Warner Telecom Inc. (Nasdaq: TWTC) VOIP executive threw a bit of cold water on the excitement over VOIP at a panel here today, saying that VOIP architectures are more complex, expensive, and difficult to operate than the old TDM systems... and less profitable (see Does VOIP Business Add Up?).

"Everybody thinks VOIP is as simple as putting a server in a rack and turning it on, and it's just not that simple," says Earl Turner, Time Warner Telecom’s senior director of VOIP technology (see TWT Narrows Q1 Loss). “I know this because implementing new VOIP networks has been my life for the last ten years.”

Because there are so many different equipment providers selling individual VOIP point products, building and managing VOIP networks gets very expensive, Turner told Light Reading during post-panel chit chat (see Time Warner Talk Fuels Sonus). “When you start looking at the total cost of ownership of the network, you see that it is actually far more expensive than TDM networks...

“What took 120 years of switching in the TDM world, we are trying to do in two to three years."

Turner said such thoughts can make him so gloomy that he's even considered a career change. “The only people that are going to make money out of these [VOIP] networks today are the systems integrators. If I would leave my job today, I would go and be a systems integrator.”

Turner spoke at a panel titled “VOIP Networks and Services” on the opening morning of Supercomm. His company, he says, is primarily in the business of providing VOIP networks and services for enterprises and institutions (see Time Warner Telecom Touts Wins and TWTC Offers VOIP Business Services ).

Despite such concerns, there is certainly no going back to the TDM world, Turner and panelists made clear. Operators live in a world where they can’t order a Class 4 or Class 5 TDM switch without giving vendors three months notice, a special order, and some explaining. “VOIP is here to stay. I’m a realist; it’s not negative or positive, it’s just what it is today.”

Not that Turner sits around pining away for the days of TDM. He believes operators will eventually begin to make money on their VOIP networks when all the pieces are in place. “The key reason that these networks are being built now is to provide a framework for more advanced services in the future,” Turner told Light Reading.

To that end, Turner urged the 20 or so service provider people in the audience to pay close attention to the services that SIP and IMS technology will enable in the future. “SIP just amazes me -- I’ll bet you five years from now even your fridge will be controlled by SIP.”

Turner advised service providers to begin engineering their networks to dovetail with the IMS architectures that he believes will dominate the future. The next-generation services those technologies enable, Turner believes, will eventually make VOIP networks a solid revenue generator for operators.

But in the near term, operators will suffer the growing pains that new technology brings, Turner says, and they will learn that it is no easy matter. The Time Warner exec says reengineering the telephony network for VOIP is a process of disaggregating the centralized method of switching that exists in TDM architectures and “putting it all back together again.” (See Session Controllers Storm Chicago.)

— Mark Sullivan, Reporter, Light Reading

<<   <   Page 2 / 6   >   >>
voyce_overipee 12/5/2012 | 3:12:10 AM
re: Telecom Exec: VOIP Is a Pain Geoff, just to correct a few errors...

The firewall traversal issue with SIP is kind of based on poor design of the protocol, and of timeout interaction between the firewall and the softswitch. The firewall has no way to identify SIP traffic (mistake in the protocol) so we have to create GǣpinholesGǥ through which SIP sessions can pass.

It's not SIP the friewall can't identify - it's RTP. SIP (and H.323) use well defined ports, but the RTP media uses dynamically decided port numbers (this is not a protocol "flaw", it's unavoidable, since you can have any number of media channels from a single ip address). So the frewall won't open those "holes", and the NAT will be modifying their address and port numbers so it breaks. Inside SIP, and H.323, they have ip addresses for things like the RTP media address and port, and some other headers. The NAT won't touch those so they end up being private bogus address so it breaks. Of course there are NATs that do understand sip or h323 enough to "fix" it locally, but it's still not common.

So why donGt H.323-based VoIP systems suffer from this problem? Well like SIP traffic, firewalls canGt identify H.323 packets easily so the protocol designers kind of screwed up there too. But most edge routers today support Application Layer Gateways for H.323 which take care of the pinhole problem.

H.323 systems actually suffer the problem more than sip, because even fewer NATs understand h323, and it runs on both udp and tcp with two port numbers and tcp connections for h245 may be created during the call setup, and is usually considered more complicated and protocl heavy with more interop problems (particularly supporting v1, v2+ and with both slow start and fast start). I don't know of a single edge router that supports h323 ALG and fixes the pinhole problem. name one.

Finally, regarding CALEA, you said:
Even Skype traffic can be snooped :-)

Actually, not from what i hear. I've been told skype encrypts their traffic from node to node, and it's not SIP or h323 or any standards'based protocol. but i haven't sniffed it myself.
Also, CALEA is not about just snooping voice traffic. You have to be able to copy the call signaling and media, of only the user under order, in a fairly secure and controlled manner. They have to be sent to specific govermentally controlled servers, and the identity of the user under warrant has to be fairly tightly secured. It's not just port mirroring and a sniffer, just like 911 support is not just a different call route direction.
optoslob 12/5/2012 | 3:12:09 AM
re: Telecom Exec: VOIP Is a Pain My short list of VOIP advantages

1)Call finds me regardless of where I am, without requiring me keep updating call forwarding or accept any bogus cell-phone international roaming conditions.

2)VOIP is inherently mobile, I can change my office location within a building without any system/PABX updates being necessary.

3)With Skype I can see the current status of my contact list and call when I see that someone is on-line. This is extremely important because I travel very frequently, as do most of my friends / contacts. So we call each other only when we see that the other person is on-line, sometimes I get calls at 3am which is OK if I'm awake and working (damn jetlag) but NOT Ok if my wife has to get up to answer our phone, only to tell the caller that I'm in China!

4) file transfer is easy as part of a call, no need to mess around with separate email addresses. I can easily copy and transfer URLGs this makes VOIP a more integrated tool.

5) calls are easy to encrypt or even to double encrypt and pack with extra random data, got to make the snoops work for their crumbs of information, very important in countries where you know that your calls are monitored.

6) Great way to bypass Hotel phone charges

7) Voice quality is often better than POTs

8) Party-line / 3 way calling is possible along with a kind of office chat come virtual bar-room. This type of system is used with many on-line games where the voice channel is often a secondary function

aswath 12/5/2012 | 3:12:08 AM
re: Telecom Exec: VOIP Is a Pain re. Msg. 13:
Its use of IP addresses inside the application protocol is atrocious...but IP addresses don't belong inside applications. (Mumble about nonadjacent layers, etc.)

I am not sure that this is a correct characterization. For example, Q.931 also puts nonadjacent layer information (channel id) in the signaling protocol. In both the cases, the entity uses the locally avialable information. But unlike Q.931, the entity that breaks the validity of the information (switch in the case of Q.931 and NAT in VoIP) does not map the information.

In any event the TW executive's lament is not regarding architectural matters. In my opinion, he is complaining that components are being supplied by multiple vendors requiring a long testing cycle. For that I say that the carriers are the ones who broke the old single vendor model. They may be justified, but then this is a price. (I think it is a bargain.)
alchemy 12/5/2012 | 3:12:08 AM
re: Telecom Exec: VOIP Is a Pain Geoff Bennett writes:
Most of us who've dealt with the equipment side, or have spoken to carriers also know that OpEx is higher for VoIP. The reasons for this revolve around the fact that IP was never intended to be either a real-time transmission mechanism, or a carrier-grade architecture. To make it perform these unnatural acts costs money, and takes time to get right. A good example is the fact that a "carrier grade" VoIP network needs one or more Border Session Controllers to overcome the gaping chasm of inadequacy that is the SIP architecture. These things don't come cheap, and they add a lot of complexity to the network design.

I think there's a big difference between "VoIP" and Cable VoIP. In the cable space, the endpoint device is embedded into a cable modem so there's no NAT traversal. It has a manufacturers certificate that is registered with VeriSign and can be provisioned in a secure way. The VoIP endpoint is "dumb" and uses MGCP signaling and a master/slave model rather than SIP. There's admission control for QoS signaling at the CMTS. There is no reason to use a session border controller in this environment.

In 3G wireless, the same sorts of arguments hold. The cell phone has a SIM card in it. It speaks a stripped down variant of SIP that doesn't need to be policed for hacks. It's not quite as dumb as MGCP but it's close. There's no NAT traversal. The radio network performs admission control for the media stream. Once again, no session border controller is needed.

I agree that you need an SBC when you're talking to a vanilla SIP endpoint to aid in NAT traversal and do functions like admission control, protocol policing, and CALEA. That's not what the cable or (soon) cellular operators are deploying.

The cable access network in DOCSIS has very good QoS with good delay and jitter attributes. The CMTS that drives all those cable modems can be a carrier-class box. As long as you locate the media gateways and soft switch fairly close to the CMTS, you can do a pretty good job of approximating carrier grade on the routed IP network (the metro part of it) without breaking the bank.

The problem at Time-Warner is that they were the early adopter. They're trying to run carrier grade VoIP on old Cisco CMTSs that don't implement the latest (and much better) DOCSIS standards and aren't carrier grade. They've adopted a model where they centralize their VoIP services. Their core network isn't carrier grade so they have outage problems. No telco is going to run a large city from a remote Class 5 office hundreds of miles away since they can't achieve reasonable availability numbers. Time-Warner does that all the time. The failure group size at Time-Warner is also huge so they need all kinds of expensive 24x7 staff on hand to bring their network back to life if there's any failure. Where they don't have the internal resources, they out-source to their equipment vendor and pay dearly for it. In my opinion, this isn't a problem with cable VoIP, it's the problem with network architecture and vendor lock at one particular service provider who doesn't have carrier-grade solutions.
fgoldstein 12/5/2012 | 3:12:08 AM
re: Telecom Exec: VOIP Is a Pain This is a great thread. The OP is wonderful too -- somebody finally recognizes that IP is not necessarily simpler than POTS.

<set ago="" few="" machine="a" wayback="" years="">
Hey, did you hear about the new Apple OS, MAC OS X, which uses a Unix-derived kernel? Now people don't have to learn that difficult Macintosh UI; they can just use simple Unix commands!

<end sarcasm="">

As I've noted before, VoIP is not one thing; it's a broad family of technologies that share one detail, the encapsulation of voice bearer traffic inside packets that are nominally IP-routed. Not even the IP header is constant (because of compression), and there are a heap of variants. Some are better than others.

Aside No. 1: SIP is clever, and has interesting long-term architectural potential, but today it's more religion than practicality. Its use of IP addresses inside the application protocol is atrocious. Yes, I know it's done because your average telephone set isn't in the DNS (though that can be fixed), but IP addresses don't belong inside applications. (Mumble about nonadjacent layers, etc.) This mistake was first done in FTP, but that was as workaround for some early 1970s gear, and when NAT happened, the conflict became obvious.

Aside No. 2: Most of the "features" of VoIP, such as those that Optoslob mentioned, are not features of IP encapsulation per se, but of enhanced signaling. CIT with TDM can do many similar things. Using VoIP to do POTS service is a different matter. Most people just want POTS-style service.

Now it is possible for a cableco to do PacketCable without worrying too much about VoIP issues -- it's a private network, and they could buy the Call Agent and Media Gateways from one vendor. If it's not Cisco the vendor might even call the combo a "switch"! (I'm thinking Tekelec, Lucent's Telica line, Metaswitch, etc. Cisco likes its own language, in which "switch" is a dumb L2 Ethernet thing and the CA and MG are separate thangs that work together but are not covered by one term.) Adding real IP networking between switch components adds lots of complexity. Distributing call agent functions is even more complex. (Media Gateways distribute nicely, provided the bandwidth issues are addressed appropriately.)

The bottom line is that network architecture is not trivial, and adding VoIP adds lots more room to make mistakes. It's one thing to have cute demo-ware or toll-avoidance-ware, and another to achieve POTS-like reliability. It can be done, but it's not half as easy as the pulverites make it sound.</end></set>
fgoldstein 12/5/2012 | 3:12:06 AM
re: Telecom Exec: VOIP Is a Pain I am not sure that this is a correct characterization. For example, Q.931 also puts nonadjacent layer information (channel id) in the signaling protocol. In both the cases, the entity uses the locally avialable information.

That's not an accurate characterization of Q.931. That's an out-of-band signaling protocol, layer 3 of a 3-layer stack -- and thus the application. It is managing channels as its application. It does exactly what it was designed to do, so there's no violation.

IP-style stacks, be they TCP/IP, DECnet, OSI, or something else, are inband. Layers encapsulate each other and exchange information with adjacent layers. What are hard-and-fast Rules of the OSI programme are ignored in TCP/IP -- not necessarily a good idea, even though it's conceded (no flame wars please) that the OSI programme went awry early on and flopped. Not because it didn't allow non-adjacent layers to use each other's information.

Now we have two plausible and non-mutually-exclusive problems with T-W's VoIP. One is the multi-vendor issue: Multi-vendor compatibility is precisely the reason why there are architectural standards. If different vendors don't agree on every last detail, or on how to deal with disagreements, then incompatibilities are possible. VoIP systems do have lots of vendors, and lots of variants. In that regard it is reminiscent of ISDN. (And PacketCable is, perhaps, reminiscent of Euro-ISDN, a tighter spec imposed by an authority that wanted more compatibility.)

Two is what alchemy mentioned -- it's easy to build a VoIP system that is not carrier grade, and not always obvious until too late. Again it is architecture, but it's applied architecture (what an SI does), not protocol architecture (what IETF types do). Lots of folks will sell snake oil solutions.
aswath 12/5/2012 | 3:12:04 AM
re: Telecom Exec: VOIP Is a Pain Msg. 16:

If Q.931 is out-of-band, then why not H.323 or SIP? After all the messages could be destined to a point different from where the media traffic is destined.

Will the signaling protocol of BISDN be an in-band? If not can you please explain?

60 Hudson St 12/5/2012 | 3:12:03 AM
re: Telecom Exec: VOIP Is a Pain Right on! to everyone on this thread.

In college I worked for a Printer. They printed 4 color separation, web (before www), off-set and litho. They had an entire department of people (about 15) who were typesetters and proof-readers. And then two things happend. The color copier and the Mac PC with spell check. One young lady fresh out of Drexel U with her Mac came in and wiped them all out. It happend in about 6 months. They were gone. That occupation had been around since Gutenberg - 550 years. Change happens. Yes, there was new money spent on new equipment that left other "paid for" equipment to rot, but hey, that's business. For the spell checkers there was no more money to be made in proof-reading. The owner of the printing company probably thought it was a "pain", but ignore it and perish.

Today I am involved in the physical layer and see a lot VoIP Peering between my customers. It's happening again and it's not gonna be pretty for many of the legacy telco's (or the States that derive most of their revenue from taxes on telco property).

OldPOTS 12/5/2012 | 3:12:02 AM
re: Telecom Exec: VOIP Is a Pain To quote Hunter GǣRight on! to everyone on this thread.Gǥ

But hunter, my daughter works for a major printer that does those ads and she does use those Mac PCs to do the layout and the color copier to test/ prototype. But they print the ads using those updated/computerized printing techniques you mentioned, because the customers demand the quality for their communications.

fgoldstein hit it on the head when he said Gǣit's easy to build a VoIP system that is not carrier grade, and not always obvious until too late. Again it is architecture, but it's applied architecture (what an SI does), not protocol architecture (what IETF types do). Lots of folks will sell snake oil solutions.Gǥ

GǣTheir core network isn't carrier grade so they have outage problems.Gǥ
Both fgoldstein and alchemy eluded to distributing media gateways for VoIP in the cable networks, but eluded to a core, leaving me somewhat confused. Here is why, I assume the cablecos MGs are primarily attached to Telcos to transport the voice between metro areas using POTS (trunks).
Is the cabelcoGs network core defined to be from aggregation points to the networks head-end or is it to interconnect the head-end?


BTW After all my Comcast complaints, T-W is swapping Comcast for my area! #$%&*()#$%&*()

jpendleton 12/5/2012 | 3:12:02 AM
re: Telecom Exec: VOIP Is a Pain Right now, the real advantage to customers of Time Warner, or any VoIP SP, and what will sell VoIP, and thus determine whether VoIP or TDM GǣwinsGǥ, is Gǣfree long distanceGǥ This results from current Federal regulation, that allows VoIP to not pay the local transport companies (LECs, CATV, or WLL) the prices they charge for use of their pipes (access charges in LEC-talk), while TDM voice calls must pay this. It is currently state of the art 1984, by the minute, so the TDM carriers charge their customers by the minute, too.

This is justified now because VoIP is a new technology, and is provided by new SPs. The TDM infrastructure (RBOC) providers may not do a good job of service, timely new feature introduction, or user control of tier service (which they pay for, but they do an excellent job of throwing lawyers at anyone who tries to compete with them. CATV, which only allows you to buy their VoIP in may cases, is not much better. To overcome this, the Feds have given the new SPs a competitive advantage until they ramp up and can really compete.

Once that happens, the Feds are going to have to either deregulate TDM or regulate VoIP. (Canada just regulated VoIP a few weeks ago, interestingly to prevent the RBOC from subsidizing their VoIP offering from monopoly TDM revenues and offering VoIP service below cost, thus driving out all their upstart competitors (I told you these guys were deviously clever).

Once VoIP and TDM voice are on the same regulation, then the GǣwinnerGǥ will be the SP that has the lowest underlying fixed network cost and the best features. TDM is still far form dead here. TDM can do Gǣfollow meGǥ, the cell guys have been dong it for years using TDM and SS-7.) And TDM could easily be modified to provide integrated messaging, computer screen web-based access to information, etc. The reason it has not been done yet is the dominant TDM providers can get as much money without doing it. Why serve the customer when you have a monopoly? Now, if we can just keep the VoIP providers from acting like the phone company, we will be set. We didnGt do a very good job of that with Internet. Once the pipe guys got us trapped with cable modem and DSL, the Internet (at least transport) acts more like Ma Bell than phone acts like you local dial-up, ca. 1999 ISP (did).
That rant aside, the future of VoIP vs. TDM for voice will be a business issue more than a technical issue. (Long live BetaMax!!! It really was better!)

<<   <   Page 2 / 6   >   >>
Sign In