328 20 NETWORK MYTHOLOGIES 20.1 INTRODUCTION As we reach the end of the book, it is useful to consider the veracity of some of the common old wives tales that abound in the packet radio arena. These pronouncements are often accepted as gospel when they are anything but. A realistic appraisal of some of these claims will surely help one make more hard-eyed, useful business decisions. Given the state of the wireless data business, a little realism will be a refreshing conclusion. 20.2 CDPD IS AN OPEN STANDARD; ARDIS AND BSWD ARE PROPRIETARY The claim is ubiquitous. AT&T Wireless says: AirData is based upon an open, nonproprietary standard. 1 Echoing this theme, Bell Atlantic claims that AirBridge Packet was designed to provide . . . (an) Open Industry Specification . . . [with a] large variety of vendors designing and building products . . . [so that] end users and customers have a wide variety of choices. . . . Unrestricted competition fosters competitive prices. 2 The CDPD Forum was slightly more nuanced: The open CDPD Specification means that multiple equipment vendors are able to provide equipment. . . . The . . . user is not restricted to a small number of licensed manufacturers. 3 Consultants are happy to pile on: CDPD was offered [as] an open specification, permitting anyone to design and manufacture subscriber and/or infrastructure products. 4 We must ask ourselves two questions concerning these claims: The Wireless Data Handbook, Fourth Edition. James F. DeRose Copyright © 1999 John Wiley & Sons, Inc. ISBNs: 0-471-31651-2 (Hardback); 0-471-22458-8 (Electronic) 1. Are they uniquely true? 2. If so, have they had the desired result? The answer to the first is: Generally no. Mobitex has been an open, nonproprietary protocol for user devices since at least 1988, before IBM even thought of proposing CDPD to McCaw. In 1989two years before I had heard of CDPDI received my first terminal specification manuals for the 8-kbps North American system. Mobitex specifications are available to any manufacturer desiring to produce modems or terminals or any programmer wishing to write application software. There are no license fees. Worldwide standardization and compatibility are handled by MOA, the independent Mobitex Operators Association. Development of infrastructure, however, may well be a sticking point with Ericsson. The DataTAC (Motorola) situation is now nearly identical to Ericsson. After mistakenly (and stupidly) stonewalling CDPD in 1991, Motorola moved to an open standard, similar to the Mobitex approach in 1992. The specifications have no license fee, but there is a caveat. Assume I am licensed and discover some neat trick while making a device for, say, RD-LAP. I must share this intellectual property with Motorola, who has shared theirs with me, on the same no-charge basis. The DataTAC specifications are managed by WWDNOG, the Worldwide Wireless Data Network Operators Group. The published WWDNOG objectives 5 are at least as sweeping, arguably even more open and nondiscriminatory, than CDPDs. The answer to the second questionDoes CDPD have more hardware providers than ARDIS or BSWD?is no. None of the main service providers have a plethora of manufacturers building for them. Recalling Chapter 16, there are basically only seven (some shaky) radio modem vendors for public packet systems. Three (NovAtel, Sierra Wireless, Uniden) deliver products for CDPD. Three (Ericsson, Motorola, RIM) equip BSWD. Two (Motorola, RIM) concentrate on ARDIS. Metricom takes care of Ricochet. There are a number of Australian, European, and Korean manufacturers of DataTAC radio modems, but their concentration on RD-LAP to the exclusion of MDC4800 make them of little value to ARDIS. The real value of multiple hardware suppliers is competitive pricing. But the high price leader in packet radio modems is Sierra Wireless. No small, relatively inexpensive device such as RIMs I@P can be contemplated for CDPD. The practical worth of this CDPD claimzip. 20.3 ARDIS HAS LIMITED CAPACITY This claim has been made by BSWD for years. 6 It may well be true in some obscure backwater where BSWD has 10 lightly loaded (even unused) half-wide channels and ARDIS has only one full-width channel. But that is just the location where there is no real capacity problem. In intense urban areas such as New York City, ARDIS matches BSWD in capacity potential (see Section 12.5 for details). BSWD might be better served to give thoughtful attention to CDPD. Without massive infrastructure investments, CDPD really has capacity constraints. Even the 20.3 ARDIS HAS LIMITED CAPACITY 329 use of TCP/IP works against it. You simply have not lived until you have tried to use CDPD in Connecticuts Gold Coast during the rush hour. Will this CDPD constraint be solved by taking over voice channels? We will soon see. 20.4 ARDIS IS OLD TECHNOLOGY This is another charge levied by BSWD in its customer presentations. It is absolutely true that MDC4800 is a 1982 design. Its bit rate is slow, and its efficiency is low. But MDC4800 is on the way out. During the third quarter of 1998 ARDIS had 410 operational RD-LAP sites positioned to reach more than 50% of the business population. Installation of the new, faster, more efficient base stations was underway in 11 additional metropolitan areas for fourth quarter 1998 operation. Of the top 25 CSMA/MSAs only Sacramento (area 25) did not yet have RD-LAP, but Portland-Vancouver and Cincinnati-Hamilton (areas 26 and 27) are up, 6 of 10 MSAs in the 30s, some in the 40s, 50s, and 60seven Sarasota, Florida (area 123) are also up on RD-LAP. The oldest MDC base stations are being scrapped (and cannibalized), with ~200 being replaced with RD-LAP infrastructure in just the Greater New York City and Los Angeles areas. In truth, the oldest technology is now Mobitex, an ironic twist on the BSWD marketing claim. 20.5 BSWD HAS INFERIOR IN-BUILDING COVERAGE ARDIS repeats this claim almost as a mantra: ARDIS Value: Deep In-Building Coverage. 7 In the past the assertion was unmistakably true in those locations having multiple, supporting base stations. But in areas that BSWD has built out for two-way paging, the in-building advantage has disappeared. It is possible to match the ARDIS single-frequency reuse penetration capability with sheer numbers of base stations. There are indications that in the northeast United States, at least, BSWD has begun to do just that (see Section 12.4.3). 20.6 PACKET PROVIDES FASTER ACCESS THAN CIRCUIT BECAUSE IT IS ALWAYS CONNECTED According to NYNEX promotional literature: Best of all, CDPD is easy and efficient because it is connectionlessthere is virtually no setup time. Once youre on the system there is no need to dial-in to send or receive a data call. This is particularly useful for those . . . who need to remain in constant contact . . . and dont have time to repeatedly dial into the network. 8 Right! If the device you are using has a car battery. Most devices used with packet systems have inadequate battery life. Some fully charged batteries on CDPD external modems will support only an hour or two of continuous activity, hence end up being plugged into the cigarette lighter all day. If 330 NETWORK MYTHOLOGIES you want anything approaching a full day of pedestrian activity, you must carry two to three sparesif the device has a pluggable replacement battery! Short battery life forces you to change your behavior significantly. . . . Rather than leaving the unit on all the time so that it can receive mail on the fly, there is a tendency to turn it on and off at regular intervals to do mail checks. 9 This is a usage profile more suitable for circuit, not packet, switched data. CDPD registration times tend to be long, varying somewhat by device. Just think how long it takes your application to come up even if you are using an IBM laptop with the Suspend feature. Many laptops do not have this capability, and the loading of TCP/IP stacks, modem microcode, and application icons can easily take two minutes, often slower than circuit switched dial access time. The source problemunacceptable battery lifeis decidedly worse on CDPD than ARDIS. The PM100D can survive a long day powered up if 9-volt alkaline batteries are used. The I@P can go a few days powered up, with recharge coming from much cheaper AA batteries. The king of battery performance, however, is BSWD, whose battery-saving protocols extend I@P950 power on time to 500 hours. 10 With that kind of battery life, one really can leave the device turned on all day, something that is just not possible with CDPD. 20.7 CDPD HAS HIGH USER BIT RATES 20.7.1 Big Claims and Shrinking Claims Glossy Baseline : AirData is fast19,200 bits per secondthe fastest wide-area wireless data network on the market. 11 Uninformed Technical Limit : The net bandwidth available to users is 14.4 k/bit/s after overhead is accounted for. 12 User Qualifier : Rates of 11 kb/s to 12 kb/s were achieved when CDPD was tested in the San Francisco Bay area. 13 Informed Technical Estimate : CDPDs . . . continuously transmitted [outbound channel] . . . yields a net rate of 9.618 kb/s. 14 Random Measurements : (Using the Sierra Wireless AirCard) it took nearly 7 minutes to download a 414K file via CDPD. 15 This equates to an effective data rate of ~7.9 kbps. 20.7.2 Combinatorial Pinging: A CDPD Static Bit Rate Test We know that AT&T Wireless fastest on the market claims are inaccurate; ARDIS brought RD-LAP up operationally in Las Vegas before AT&T could demonstrate CDPD at Comdex. But when will the shrinking stop? Who is right here? How can the user do independent tests, without special equipment, to form a separate opinion? To obtain some insight, CDPD tests were conducted under extremely favorable conditions: The device was static, an external modem was mounted high in a building 20.7 CDPD HAS HIGH USER BIT RATES 331 with direct line of sight to a nearby base station, the received signal strength indicator was never less than 69 dB, and the transmit power never had to go beyond level 2. There is no evidence the radio environment was ever hostile. In the evening, assured by the carrier that other test users had left the system, simple static stress testing was performed with multiple subscriber units. Each of the devices were used to ping the same base station at 1-second intervals with a 2-second time-out. Note that the Chameleon ping routine is not precise about time. If one specifies 2 seconds as a permissible time-out, failures will begin to occur at some time greater than 1.8 seconds. The packet loss = 0 point occurred with all three units running 300-octet messages. The statistics are: minimum time 0.933 seconds, average time 1 seconds, and maximum time 1.263 seconds. Since ping messages have only 32 bytes overhead, this is an effective combined outbound data rate of 996 octets per second: 7968 bps. This is remarkably close to the figure reported in the random measurement of Section 20.7.1. User messages have more address overhead than pings, but a best-case rule of thumb of ~8 kbps probably represents an upper limit for CDPD under ideal conditions. Try it yourself at 3:00 AM with your own TCP/IP stack. 20.7.3 Realistic CDPD Bit Rates in a Busy Channel First, how can anyone have seen an 1112-kbps rate? Answer: By sending a single, one-way, ~1375-character UDP message, without contention, and measuring just its transmission period. It may well complete in just 1 second, an instantaneous rate of ~11 kbps. But the full second consumed adds to the response time of a contending message waiting to go next. CDPD airtime protocols were designed to ensure that no single user can dominate network resources. Each carrier can set a parameter for its own system called MAX_BLOCKs. The default setting is 64, which is roughly 2000 bytes. For sake of argument, assume that the carrier has set MAX_BLOCKS to 44, and 1375 user bytes can be delivered in 1 second before control must be relinquished to the next in queue. Assume there are three users on the channel, each asking for a Web site response of several thousand bytes. Then the multisegment responses will look like Figure 20-1. Over the long haul, with a busy channel, message segments are delivered at an average rate of 1375 bytes every 3 seconds, an effective data rate to each user of 3665 bps. If you gave up on your 14.4-kbps landline modem for Web access and are barely content with 33.653 kbps, consider how you will feel at the CDPD rates. In truth, Figure 20-1 Resource competition: outbound channel. 332 NETWORK MYTHOLOGIES Web access is a lousy way to use CDPD from a users point of view. You would be better served on circuit switched cellular (or Ricochet). But the carriers would rather have three users paying for this channel than one. If you can stomach it, they will take your money. 20.8 CDPD WILL MOVE TO HIGHER BIT RATES WHEN THEY ARE AVAILABLE This is conceivable, but only if a new airtime protocoland new radio modems and base stationsare staged into place. The technical task would be akin to the ARDIS migration to RD-LAP, but on a far grander scale, especially for infrastructure. The magnitude of the effort is such that migration to something totally different, such as the winning technology resulting from the ITUs IMT-2000 initiative, is more likely. Why would a new airtime protocol be required? The answer is because of the choice of Reed-Solomon coding for forward error correction. Recall from Section 13.2.3 that CDPDs Reed-Solomon implementation takes 35 1/4 bytes, 282 bits, and converts it into exactly forty-seven 6-bit symbols. These are then error encoded to produce exactly sixty-three 6-bit symbolsan efficiency of 47/63 = 74.6%. The resulting 63-symbol block can theoretically withstand 8-symbol errors anywhere in the block. In fact, the undetected error rate is so high at this level that some carriers back off to the better 7-symbol correction. Errors can manifest themselves as 7 random hits, a burst error 7 symbols long, or any combination in between, as shown in Figure 20-2. Under ideal conditions (no random errors), a CDPD block can withstand a burst error that is 19.2 kbps 7 symbols × 6 bits = 2.19 milliseconds long This is just marginally acceptable. Recall that Mobitex (BSWD) can withstand fades of 2.5 milliseconds, and creaky old MDC4800 (ARDIS) can take a 3.3-millisecond hit. One certainly would not want to take CDPD below its current design point. Now suppose CDPDs raw bit rate could magically be doubled to 38.4 kbps, which, frankly, still is not all that interesting. If the airtime protocol is employed unchanged, the best case fade protection would fall to 1.1 millisecondsunacceptable! The Figure 20-2 Reed-Solomon error correction: CDPD. 20.8 CDPD WILL MOVE TO HIGHER BIT RATES WHEN THEY ARE AVAILABLE 333 protocol, the base stations, and the user radio modems must be changed to accommodate the new error correction scheme. That is truly unlikely. 20.9 ARDIS, BSWD, CDPD (PICK A NAME) WILL BE OUT OF BUSINESS SOON This claim, referring to ARDIS and BSWD, was first floated in 1991 by members of the carrier consortium that had signed up UPS for a circuit switched cellular solution. The predicted end date was 1992. Now UPS has gone to ARDIS, and each of the carrier presenters have long since left the wireless data business (some with temporary stopovers at CDPD). Recall the consultant who predicted that ARDIS and (BSWD) . . . will reach their peaks in total subscribership in 1996 . . . then will suffer rapid business declines . . . and go dark around 2000. 16 Instead, both carriers are having a fine 1998. In the past BSWD has hinted darkly that ARDIS has an uncertain future. 17 Dont we all. But in the first quarter of 1998 venture capitalists thought prospects good enough to lend AMSC $335 million to fund the . . . growth needs of the combined company. 18 Now it is CDPDs turn: [The] window of opportunity is closing at the end of [1998]. 19 Or integrating data into a voice phone is a very good idea, but I believe the idea will succeed in digital cellular and PCS rather than CDPD. . . . CDPD . . . will simply fade away as other networks capture the customers, such as RAM, ARDIS, digital cellular, PCS and Nextel. 20 There is no question that wireless data is a vicious, tough businessand there have been more than enough casualtiesbut it is a bit early to call it quits for these three players. 20.10 WHEN WIRELESS DATA SUCCEEDS, IT WILL BE A GREAT BUSINESS! The one sure consequence of this perception has been to draw in too many players. The wreckage of failed attemptsfrom CoveragePLUS to Geotekis stark evidence of how hard this game is. The surviving players engage in fierce price battles to win market share. At these cost points, is there any profit left for anyone? 20.10.1 ARDIS Before its acquisition by AMSC at the close of 1997, ARDIS claims to have reached the break-even point. But new high-traffic wins at customers such as UPS, and the stinging realization that BSWD had clearly improved its in-building coverage in key metropolitan areas, kicked off a new round of infrastructure expansion. If we could examine just ARDIS, free from its money-losing parent company AMSC, how near is ARDIS likely to be to break even in 1998? 334 NETWORK MYTHOLOGIES A key expense parameter is the amount of infrastructure employed. ARDIS had 1750 base stations at the close of 1998. About 1000 were DCS vintage, installed in 19831985, and are long since amortized. But the balance is much newer, with a rich RD-LAP mix. To simplify the case, assume that the total cost to buy and install a new base station is $60,000. Assume further that an amortization interval of 10 years can be justified. The monthly amortization per new base station is $500; the annual amortization cost for the 750 new base stations is $4.5 million. Unfortunately, equipment amortization is just the beginning. All 1750 base stations must be maintained, at an annual cost estimated to be ~10% of the original purchase price. This is a nontrivial expense of $10,500,000. Until recently, each base station had to be served by a unique leased analog line. With modems, and the prorated share of the AT&T licensed space arrangements, it was not out of line to say that each base station cost $500 per month in Telco costs. To reduce these charges, consolidation of multiple base stations at one physical location permits the judicious deployment of T1 lines, and even frame relaythough single-frequency reuse timing constraints and dial backup can be a problem with the latter. The average monthly line costs are now estimated to be $400 per base station. Each base station sits on a rooftop. Some of these are very rural and inexpensive. But most are in premium metropolitan areas commanding top dollar. Further, even the rural rooftop/backyard owners have begun to realize that antenna space is worth money. Assume that the average monthly rooftop cost per base station is $400. ARDIS also has a payroll to meet. Smaller than its competitors, this nut is not out of sight but must be at least $30 million, including benefits. Missing from the equation is the cost of buildings, computers, test equipment, and so on, at Lexington and Lincolnshire. We will add a 20% contingency to pick up at least some of the missing items. These figures mean that ARDIS needs the following annual revenue to break even: Wages, salaries $30,000,000 Contingency (20%) 6,000,000 Base station amortization (750 × $500 × 12) 4,500,000 Base station maintenance (1750 × $6000) 10,500,000 Leased-line costs (1750 × $400 × 12) 8,400,000 Rooftop rentals (1750 × $400 × 12) 8,400,000 $67,800,000 We know from the AMSC quarterly report 21 that ARDIS average monthly revenue per subscriber dropped to $52 in the third quarter of 1998, down from ~$75 in 1996. To reach break-even, ARDIS must have $67,800,000 $52 × 12 = 108,653 subscribers AMSC actually had 101,500 subscribers as of September 30, 1998. Of these, 12,400 were voice users, a slow-growing category not germane to ARDIS. The 89,100 data 20.10 WHEN WIRELESS DATA SUCCEEDS, IT WILL BE A GREAT BUSINESS! 335 subscribers contain a segment that is satellite only, no ARDIS involvement at all. It is probably safe to say that core ARDIS had 80,000 subscribers at the close of the quarter and might approach 85,00090,000 subscribers by year-end. Core-ARDIS will fall short of break-even in 1998. The coming of UPS quantities in 1999 will drive the subscriber counts well above the target, but there will also be additional base station costs to absorb. Nevertheless, ARDIS will flirt with the break-even point and can be visualized as going into the black, on a period basis, at the millennium10 years after its beginning. 20.10.2 BSWD Let us project ahead to the 2500 base station goal BSWD has for two-way paging. Mobitex base stations are often multichannel and more expensive than the ARDIS single-channel units. There is also little, perhaps no, colocation. For this analysis an installed cost of $70,000 per base station with a 10-year amortization period is assumed. The BSWD wireline back-haul problem can actually be more difficult than ARDIS. But some base station line costs will likely be absorbed by parent Bell South mobility in areas where it has cellular service. Thus, a monthly leased line cost of $400 is assumed. The BSWD payroll is larger than ARDIS, with 330 employees in February 1997. 22 The 1998 payroll cost is almost certainly greater than $35 million. BSWD has always priced aggressively; its drive toward two-way paging is certain to keep the average revenue per subscriber down. We will assume $48 per month, the price for the (barely) discounted 400-message, two-way pager plan. 23 The BSWD annual costs look something like this: Wages, salaries $35,000,000 Contingency (20%) 7,000,000 Base station amortization (2500 × $7000) 17,500,000 Base station maintenance (2500 × $7000) 17,500,000 Leased-line costs (2500 × $400 × 12) 12,000,000 Rooftop rentals (2500 × $400 × 12) 12,000,000 $101,000,000 The subscriber break-even point is thus $101,000,000 ( $48 × 12 ) = 175,347 BSWD may close 1998 at ~50% of that mark and will need three to four more growth years to approach the break-even point. Another approach is to charge off much of BSWDs cost to its own field service users. This is good book-keeping, but not necessarily good business. 336 NETWORK MYTHOLOGIES 20.10.3 CDPD The cost is truly trivial, said some CDPD proponents. 24 But this is simply not true. The 1994 average cost to install a CDPD base station was $50,000 to $75,000 per cell site. 25 We know every cell site in the nation will not be fitted out for CDPD, but suppose the four leading carriers try to convert 30,000 cell sitespossibly half the current nationwide total. Then the combined carriers will face a capital outlay of $1.5$2.2 billion . That is unlikely to happen in a world where money must now be allocated for digital cellular and PCS. Thus, CDPD is likely to be employed only in regional areas where there is a reasonable shot at (someday) making a period profitnever mind the lost years of earlier investments. The Greater New York City CSMA is one such area. With a population approaching 20 million8% of the nationit is not too difficult a task to project a pool of ~300,000 potential two-way wireless data users. Some of this potential will be picked off by circuit switched cellular, some by ARDIS and BSWD (both very strong in field service, about 89% of the potential), some by SkyTel2, some by Nextel, some by Omnipoint, but the pool is large and there should be some left for CDPD. To compete, the two leading carriers must at least complete the infrastructure rollout, and dedicate the necessary channels, in the business area. There also may be unusual business arrangements. AT&T Wireless is unlikely to leave Southern New England to SNET and may strike some sort of CDPD lease arrangement with SNETs new owner, SBC, which will become a CDPD participant with the acquisition of Ameritech. To estimate the size of the required investment, we will examine just the five boroughs of New York City and the surrounding counties of Connecticut, New Jersey, and New York State. This will be limited. We will exclude counties such as Monmouth in New Jersey, areas of Orange and Putnam Counties in New York, and parts of Pennsylvania, even though they are included in the CSMA definition. To estimate the number of cell sites, we will divide the geographic area by the cell coverage. The areas of the counties of interest can be obtained at a good library. 26 Cell site radii in New York City tend to be short. One need only drive the Brooklyn Heights, Borum Hill, and Red Hook sections of New York Citys most populous borough, look at the rooftops, and watch the odometer to see that a 1/2-mile radius is not uncommon. Well use 3/4 mile as an average radius for the boroughs. In the border counties we will back off to an average 3-mile radius. Thus, each carrier must erect 304 trisectored CDPD cells, as indicated in Table 20-1. This is not a fantastic number given that NYNEX (pre-BAM takeover) rolled out 176 base stations in its first wave. Drawing upon the calculations made in Section 11.4.5, each carrier needs 25.4 data users to break even with voice. For CDPD to make sense in Greater New York City, 23,165 subscribers are needed by each carrier (25.4 subscribers × 3 sectors × 304 sites). This is a major stretch and will simply not be met short term. If achieved in, say, five years, CDPD would account for ~20% of all two-way data users in the city. 20.10 WHEN WIRELESS DATA SUCCEEDS, IT WILL BE A GREAT BUSINESS! 337