A recent entry in Skype Journal talked about a device called from Broadband Booster from Hawking Technologies. It seems they coordinate the uplink traffic in such a fashion critical traffic like VoIP does not encounter much delay. From the material available in their website, I am not able to figure out how they manage it. It seems, Andrew Sheppard has done some analysis. For me, this raised more questions rather answering old ones.
According to Sheppard, use of Broadband Booster increases downlink bandwidth to 243 KBps from 227 KBps. I am not sure how this is possible. The values for send and receive packet loss are beyond my interpretation. I tried to get an explanation by posting these questions in Skype Journal. But as it has happened before, my comment didn’t make it. If you understand the report or know how the product works, please share your knowledge.
We all know that Skype uses P2P technology and heard the claim that this has allowed them to scale very inexpensively. The crux of the idea is to conscript a few, select users to act as “supernodes” and “relay nodes” that assist other users during establishment of calls and for traversing NATs. People have theorized and have collected empirical evidence that Skype has seeded the network with a few of their own supernodes. I have a suspicion that eBay may conclude, for corporate and commercial reasons, it will play safe and deploy all the needed supernodes and host them in their data centers. So I wanted to find out how expensive would that be.
Previously Zennstrom had stated that a conscripted supernode will manage about 100 clients. Tithing suggests that such a supernode will use one tenth of the computer’s resource. The modern religion, also called secular government, takes away about one fifth of its citizens’ income. So we can assume that a computer that is dedicated only to be a supernode can support 500 to 1000 users. Let us assume that the cost of a computer is $1000. Currently Skype claims it has 60 million registered users and at any time about 4 million users are logged onto Skype. If we use this ratio, then it costs around 6 to 13 cents to add a user.
We are also told that a supernode consumes 5 Kbps. I think there is a typo and that the bandwidth consumption is 5 KB/s. This is for 100 users. Hence a dedicated computer needs a net connection of 25 to 50 KB/s and will consume about 4 MB per month. I contend that even the operational cost is marginal.
In short, it is neither capital intensive nor cost much operationally to deploy and operate supernodes for Skype. More importantly, this is not unique to Skype. A SIP based system can also be deployed using a similar architecture.
Yesterday Neustar announced a new set of “application peering services” called SIP-IX. According to Light Reading, SIP-IX is not just ENUM services. It seems, “SIP application calls, or “sessions” are a little more high-maintenance.” For a SIP session to traverse from one network to another, the two networks must “negotiate a number of other SIP-oriented parameters like provisioning and OSS, SIP security, QOS, and traffic accounting and settlement.” This is according to Mark Foster, CTO of NeuStar. Dan Dearing, Vice President of Marketing, NexTone expresses similar points in an interview with Tehrani. Tehrani has also predicted that 2006 will be the year of VoIP peering. Irwin Lazar comments on these and points out that there is a mailing list discussing the issues of peering.
I am all lost. I have a rather simplistic view of the issues associated with peering. Peering is needed for two reasons – to share the directories and to carry the media traffic across the network boundary.
The former makes sense because the carriers hoard the identity analogous to email ids. Personally I prefer that IP Communications follows the Domain Name model, in the sense that my id is not determined by the service provider I presently use.
The latter is perplexing. It is not clear how many service providers offer QoS. If a service provider offers unlimited in-network calls, then one can be reasonably certain, they do not offer QoS. But if they do offer, then haven’t we recreated PSTN model of interconnecting? Richard Stastny asks the same question.
In any event, for the end users, all these activities are inconsequential. A service that participates in a SIP-based peering program might use a proprietary scheme towards the end-user (as Skype is doing it). Even if a service provider uses SIP nominally, they can make it a close system (as Vonage has done). For the end-user, a major advantage of a SIP based system is that the client can be sophisticated and exciting features will be made available to them. These are not guaranteed just because SIP is used for peering. Just look at PSTN. It uses SS7 and ISUP for interconnection, but the access could be Tip and Ring and the phone could be a rotary phone.
On Friday, Jeff Pulver posted his concerns regarding the FCC’s First Report and Order on CALEA.. He ended the post thus: “I welcome your thoughts and extension of this dialogue.” Hence this post.
The following is a quick summary of Jeff’s post:
In this post, I am not arguing that CALEA should not be applicable to IP Communications. This will be decided through the political process. Instead my focus is how should it be done IF we decide to require CALEA?
First, I want you to observe that in PSTN, all calls can be intercepted – not just voice calls. So CALEA is not just a voice requirement. Instead it is an access requirement. Then why this specific focus on VoIP?
Second, I want to bring to your attention that complying to CALEA requirement will not only be expensive for VoIP providers; it will also destroy the basic architectural advantage. As Randell Jesup observes in his comments to Jeff’s post, the only way a VoIP service provider can comply with CALEA is for them to deploy media relays and route all calls through them. This is because CALEA requires that the targets must not be able to discern that they are targets. Now we know that it is routine to use media relays to assist in NAT/FW traversal. But they are used only in the initial segment of the call. After the initial period, the media are sent directly between the end-points. For the intercepted calls, the redirection will not take place and the absence of redirection could be an indication of being intercepted. I read in a New York Times article that at most 2000 or so orders are approved for Call Content interception. Given this low number, it is a high price to route the media through a media relay for all calls. But more importantly, this reduces VoIP architecture to be equivalent to PSTN and will require the same level of effort in designing and maintaining the network.
Third, any end-point can decide to interconnect a VoIP service to PSTN. Does an extraneous act force the VoIP provider to comply with this Order?
From these points it is clear that direct translation of implementation of CALEA in PSTN shouldn’t be taken over to IP. Instead it should be adopted. A VoIP provider has access to the signaling information (part of what is called CII) and the ISP access to the media (what is called CC). But the ISP does not which specific flow needs to be intercepted. So if an intercept order grants access to CII alone, the the order will be executed by the VoIP provider. And if the order grants access to CC, the the order should be submitted to both the VoIP provider as well as the ISP. When the call is initiated, the VoIP provider will provide the IP address and port number to the LEA, which in turn will pass it on to the ISP (in real time), which can intercept the media flow accordingly.
A couple of secondary observations. Some have observed that the media could be encrypted thereby nullifying the benefit of interception. This is a red herring. In PSTN, one could use an encrypted (admittedly a weak one) phone, like STU III. That didn’t preclude from the application of CALEA.
I am a bit puzzled about the jurisdictional boundaries. Consider the case where an intercept order is issued for NJ and the target is connecting from NY. Can the call be intercepted?
It is advantageous to allocate many of the regulatory requirements between VoIP providers and ISP by looking at the PSTN model and ascertaining whether it is levied as an access service or as a voice service.
Lately, Long Tail has been referenced a lot. Today Tom Evslin has written an excellent summary of the idea behind it. As Tom points out IP technology makes Long Tail happen. But so far it has not happened to voice communication. The purpose of this note is to suggest one way.
Till now, the major source of commercial activity seems to be in interconnecting to PSTN. But, the charges vary widely and no single provider is uniformly cheaper. For example, Skype charges $0.02 per minute for calls to US whereas Gizmo Project charges $0.017; Skype charges $0.15 per minute to India, whereas Gizmo Project charges $0.23. To call India, one can use Reliance from a PSTN phone and its charges are $0.13. For comparison, AT&T charges $0.23. This suggests that each provider is getting different wholesale rate for different countries, possibly based on the traffic load.
Long Tail suggests that consumers must be able to reach the appropriate interconnect carrier for the specific destination. Of course we do this in PSTN with two stage dialing. SIP makes it easy to do in a single message exchange. But almost all clients do not have a mechanism and VoIP service providers do not allow users to connect other interconnect providers. PhoneGnome, which bills itself a product and also does not offer its own interconnection service, is a possibility. I am not sure whether currently they allow this or they limit only one interconnect provider.
SOMETIMES what appears to be a threat is actually a life preserver.The poor defenseless music industry cowered - then prosecuted - when the monster of digital downloads came lurching over the horizon. Then the iPod came along and music looks like a business again - a smaller business, eked out in 99- cent units - but still a business.
Cable channels were supposed to gut network television, but instead have become a place where shows like "Seinfeld" and "Law and Order" are resold and rewatched. The movie industry reacted to DVD's as though they were a sign of the imminent apocalypse, and now studios are using their libraries to churn profits.
Which brings us to the last of the great analog technologies, the one many of you are using right now.
From Forget Blogs, Print Needs Its Own IPod by David Carr, New York Times, Oct 10, 2005-10-12
Thus began a recent article in New York Times. As you read this, one might anticipate that the story is about the telephone network. But then it long since has ceased to be an analog technology (notwithstanding an occasional derisive comment to that effect). In any event, this started to make me think how the PSTN operators could make their offerings comparable and competitive to those that COULD BE (very few are actually offered) offered via IP. This note is an attempt to capture some of my thoughts.
The first assumption I am making is that the most of the PSTN subscribers will be comfortable to stay with the service rather than switch to VoIP if the features are comparable. For example, if a customer prefers flat-rate billing structure and a PSTN carrier offers it, then the customer will not migrate to VoIP. If a prospective VoIP customer likes the Presence capability, but a PSTN carrier offers a comparable capability, then it is very likely that the customer will stay put. In other words, the First Law of Market Behavior applies – Customers will continue to use a long used service until and unless a new offering is substantially better or different. This means that PSTN better plan on having feature parity.
For all the talk of “Stupid Network” most of the VoIP service providers are really recreating the same old “Intelligent Network”. They have to; otherwise there is no need for them. This is a big relief for PSTN operators because architecturally there is parity. So any feature distinction comes about because VoIP uses an out-of-band, message oriented signaling. If at all, POTS can offer only a rudimentary enhancement as is done for Caller ID. But this is an expensive proposition. This implies that one has to come up with a smarter way of conveying signaling messages without requiring a major upgrade to the existing system.
In the real IN architecture used in PSTN, there are two entities – SSP and SCP. The SSP is the traditional telephone switches and SCP is like an Application server where feature logic could be executed. Since the SCP is a computer, it could be augmented with an interface to the Internet and it could be used to generate and receive the message oriented signaling from the PSTN end-points. This is a simple enhancement.
The problem is not fully solved. There has to be a device at the customer side that can receive these messages and present it to the users in a meaningful way. In other words, as the title of the NYT story (Forget Blogs, Print Needs Its Own IPod) suggests that PSTN needs its own iPod. What should be the design of PSTN iPod? If we restrict the market to those who have broadband access, then think of an enhanced cordless phone system where the base station also has an Ethernet interface. This way the base station can be connected to the SCP. Also the base station can use the screen of the handset to display information like buddy list, presence and enhanced call handling of a waiting call.
As I see it, if the VoIP industry does not heed the advice of the Jedi and realize the force of IP Communications, the Empire can strike back.
There are press stories describing AT&T’s approach to address E911 requirement. I could not locate the press release at their site, but interesting approach develops if one combine news item at Reuters and at TopTechNews. The basic idea is to demand explicit identification of the location while reconnecting after a disconnect. If the location remains the same as before or if the new location is in one of the covered areas, then AT&T service will resume; if on the other hand, the area of the new location is not covered (only 50% of the nation is covered) or if the user declines to provide location information, the service will be interrupted, even though access to 911 will be provided. One AT&T spokesperson is quoted as saying that this is “not the most elegant solution” and another one is quoted as saying that “this is the best that technology can offer.” But one hopes that they continue to deliver the incoming calls. After all, AT&T benefits from terminating calls especially when the user does not make any outgoing calls. The news accounts are not clear about what happens when the user has traveled to foreign country. This sure looks like an over-reaction to a well intended and modest regulatory requirement.
Update: Andy has posted his thoughts on this matter. I would have expected that he will be disappointed with the high-handed approach taken by AT&T. It is not clear whether he realizes that he can not take his CV box to UK or France or half the places in US. We have to wait and see his reaction after he gives it a try.
In the aftermath of the devastation caused by Hurricane Katrina, Stuart Henshall proposed that displaced citizens of the affected area can be reconnected to the communications network, if their telephone numbers are “virtualized” and rerouted to them using VoIP technology. Being a splendid idea, it was immediately seconded by many influential bloggers like Jeff Pulver, Tom Evslin and others. The purpose of this note is to analyze what it takes to achieve this objective and to point out potential long term implications if we take this proposal to the limit.
I suspect that “virtualizing” a number is derived from “virtual number” service offered by many VoIP service providers. With this service, a VoIP subscriber residing in a geographical area can be reached via a number normally allocated in a different geographical area. So, in this case, the residents of Gulf coast can receive calls directed to their old number will be handed over to their number in the area where they have relocated to. Most people had tacitly assumed that this will be done by a VoIP service provider. But if you look at the technical details, it could be done by PSTN operators as well, because the fundamental technology is Number Portability.
With Number Portability, calls terminating at a switch can be redirected to another switch. Due to PSTN’s peculiar rate structure, the application of Number Portability has been restricted to only those cases where both the switches are in the same rate center. (This is called Local Number Portability.) In the case of Katrina, Stuart’s proposal would not have helped because it is very likely that all the switches in the affected area would have been in disrepair. Accordingly, FCC in its emergency ruling suspended the “local” restriction and said the number could be ported to any location. So conceivably, BellSouth could have assigned an alternate switch for all its customers. This is my first observation, but not the main one.
Due to the “local” restriction, the needed Number Portability query is postponed to the last possible switch and it is called “N-1 query”. But if we remove the “local” restriction, then it is better to perform the query as early as possible, what I call “1+ query”. On top of that if we add a further enhancement, that of the response to a query to be dependent on the source of the query. Then there is no more need get a virtual number. Instead the VoIP provider just needs to route the call to the nearest gateway. Users can derive more features along this line. If only …
Sprint Nextel announced that one of its subsidiaries has sued Vonage and Voiceglo regarding their patents on VoIP. But the curt press release raises more questions. For starters, we are not told the specific subsidiary and what are the patents (but seven of them, we are told) and the nature of their claims. So we can not answer questions like, whether this will affect the whole VoIP industry, will we be able to work around these patents and more importantly whether the claims can withstand prior art challenges.
But meanwhile, lots of attention is paid to Sprint Nextel. So one wonders what the real story here is.