Fallout from the ANS "Proposal" of January 1991

At the February 14, 1991 Internet Service Provider's Workshop held at the United States Congress Office of Technology Assessment by the author, Bill Schrader, President of PSI and Susan Estrada Director of CERFnet announced that CERFnet a California based mid- level, PSI and UUNET had agreed to interconnect their networks as the Commercial Internet Exchange (CIX) with a pledge that they would freely pass unmetered commercial traffic between each others networks. Since UUNET and PSI both ran coast to coast T-1 backbones, the CIX presented the mid-levels attached to the NSFnet with the possibility of an alternate backbone should relations with ANS become untenable.⁶⁴

As the spring and summer of 1991 passed ANS applied its strategy of attempting to pry loose clients from the mid-levels. While it met with very little success, from a succession of postings to the com-priv mail list and from numerous conversations that the author had with individuals involved among the mid-levels, it became apparent that it was creating considerable mistrust and resentment.

At the end of May it was reported that ANS was getting ready to launch a for profit subsidiary. At 12:32PM May 24 Steve Wolff gave his blessing to Eric Aupperle in the following message:

Subject: ANS commercial traffic
To: Eric M. Aupperle, ema@merit.edu
Cc: jcavines@nsf.gov, steve@cise.cise.nsf.gov
From: steve@cise.nsf.gov

This is to confirm our agreement on the flow of commercial traffic across the NSF sponsored gateways to the T3 network.

NSF agrees that ANS may move commercial traffic in both directions across the NSF sponsored Backbone gateways, providing that:

1. ANS recovers at least the average cost of the commercial use that traverses the NSF sponsored gateways.
2. Excess revenues recovered above costs for this use after tax will be placed in a pool to be distributed.

3. An ANS resource allocation committee will be formed with representation from the participating NSF sponsored gateway management, other network organizations, the NSF and ANS to distribute those funds with the objective of further building national and regional infrastructure, and

4. MERIT and ANS ensures that the attachment and service sponsored by the NSF under Merit's Cooperative Agreement with the NSF is not diminished.

NSF, MERIT and ANS will agree on the technical means of compliance with the points outlined above."⁶⁵

Launching ANS CO+RE

In the pages of the June issue of Telecommunications, Brian Kahin, Director of Harvard University's Information Infrastructure Project was quoted as noting that the "for-profit subsidiary offers a vehicle for substantial capital investment that would ordinarily be unavailable to a non profit corporation such as ANS, funds that would both come from IBM and MCI. However if ANS proceeds with this approach it will raise a host of critical and unresolved regulatory questions concerning the company's responsibilities to the public interest as a quasi-monopoly provider."⁶⁸

The July issue of Telecommunications featured an interview with Weis who said that while ANS CO+RE would allow commercial and R&E users to interconnect, "a main objective of the company will be to develop a national data network that can offer high bandwidth connections to corporate customers. . . . . Weis said that the ANS network, for example, could be used as a cost effective alternative to a nationally deployed private T-1 network. . . . . For corporate customers, ANS CO+RE Services will provide 'one stop shopping' including the deployment of routers on the customer's premises, the provision of local exchange circuits, and the planning and installation of network equipment."⁶⁹ ANS, obviously aware that the outsourcing of corporate networks was turning into a multi billion-dollar-a-year business was ready to try for part of the action.

ANS supporters asserted that IBM and MCI, by "cost sharing" were themselves contributing three to four times the amount of money received from the NSF in order to provide the NSFnet backbone services. If these assertions were correct, then the NSF and US Government was just the first customer of a data network established with ten to fifteen million dollars a year in contributions from its corporate parents. On the other hand, if the ANS Form 990 for its critical first year shows all the cost inputs, we find out that contributions from its corporate parents totaled just over $6 million while contributions from the US Government were $9,343,000.^ll A critic could look at the latter set of figures and suggest that not only was Congress acquiescing in the use of tax payer dollars to fund a competitor of PSI and Uunet, but also a would be competitor of rather larger companies such as Sprint Net and BT Tymnet.

No matter who was subsidizing, whom with ANS CO+RE required to put profits from its success into the infrastructure pool, the NSF could perhaps feel that by assisting in the development of this commercial company, it was engaged in a venture that would benefit its constituency. However apart from the question of whether ANS was making enemies within the network world, there was the question of whether, from a technology point of view, it was positioned where it could expect to succeed. For would ANS have, from a cost point of view, to position its services against the fast packet (frame relay and SMDS) services offered by the telephone companies.^mm It would also have to deal with security as another issue. George Colony, President of Forrester Research told Telecommunications that the Internet is not thought of as a highly secure network among Fortune 1000 users. "The freedom and openness of the Internet which has been its biggest benefit, is also its biggest problem," he noted.⁷⁰ With ANS positioning itself to offer turnkey outsourcing, it was offering itself as a carrier for all a corporation's data -- not just for conversations between corporate researchers and academics. Memories of the infamous Morris "worm" would have to be in the minds of those to whom ANS would sell its services.

By virtue of the proposal that MERIT had submitted to the NSF in 1987 ANS CO+RE in 1991 could differentiate among its customers only at Layer 3 of the OSI protocol stack. Real data security however would be found only with networks controlled at Layer 2.ⁿⁿ Because the Department of Energy and NASA run national laboratories where classified work is done, they - like Fortune 1000 businesses - must be very security conscious in their network development and indeed their networks with classified data are never connected in any way to the Internet.

In December 1992 their plans were outlined in the draft version of the OSTP report to Congress on NREN:

DOE and NASA will be spearheading the technology evaluation and deployment of cell relay services and protocols at the Open Systems Interconnect (OSI) Layer 2. These Layer 2 services and virtual private networks (VPN) allow for greater resource management accounting and control than typically provided in traditional Layer 3 networks. in addition DOE and NASA's choice of Layer 2 services is based upon Broadband ISDN standards advocated by the telecommunications industry as establishing the future direction for communications systems. Therefore DOE and NASA will be in full accord with the evolving worldwide telecommunications infrastructure and will be able to interoperate with the general research and education community as new telecommunications products and services emerge and become commercially available. The DOE and NASA networks will initially interconnect at the FIXes, along with the next generation of the NSFnet backbone which itself is targeting advances in routing and peering technologies at OSI Layer 3.⁷¹

A person directly involved told the author that the NASA DOE effort had to focus on Layer 2 in part because of security concerns, adding that the Fortune 1000 would want virtual private networks which could be easily run over communications facilities at Layer 2 but not at Layer 3. Business also tends to require more protocols than just IP, and a Layer 2 based network will allow them such options. A focus on Layer 3 he added, is the best choice, if your mission is to interconnect hundreds of thousands of hosts and millions of users in a freely accessible open interoperable manner using a common protocol like IP.

Planning for November 1, 1992

"the NSF may make a single award to a backbone service provider after a standards NSF competition - i.e. 'recompete the backbone'; ANS might win but so might Sprint. Or AT&T teaming with somebody, or PSI, or . . I regard this option as very unlikely to be chosen.

Or [the NSF might] allow those who need backbone services to compete for the pot of money NSF now puts into the backbone via MERIT, and spend the money with any eligible provider ('eligible' at a minimum means participating in CIX/FIX - ie, agreeing to carry competitor's traffic.)"⁷²

It should be noted that the second condition held an implied threat for ANS which insisted, because it provided the biggest and fastest backbone, it would measure and charge for carrying competitor's traffic. This was in contrast to the CIX where all members agreed to freely interconnect their networks without imposing settlement fees on a competitor's traffic. ANS was refusing to join the CIX. Wolff's plan to impose the CIX model of interconnection as a precondition for eligible for backbone attachment subsidies from the NSF should it not recompete the backbone was very likely the best way that Wolff could avoid a split in the network into competing CIX and ANS camps no longer connected at the FIXs (Federal Internet Exchange Points) when the MERIT Agreement expired in November 1992. ANS' forced march to gigabit speed was introducing all sorts of complicating factors into the NSFnet.

On August 2, 1991 the founding members of the CIX raised the profile of their proposed method of Commercial Internet formation by founding the CIX Association as a 501(c)(6) trade association where all commercial providers of TCP/IP services were pledged to interconnect with each other and not discriminate against each other's traffic.⁷³ Since, if its proposal of January 15, 1991 is to be believed, ANS needed hefty settlements revenues from traffic that crossed its network in order to achieve its goals, ANS refused to join the CIX.

Instead ANS, the mid-levels, and the other commercial providers prepared for a climactic FARnet meeting scheduled to convene on August 14, 1991 at Big Sky, a resort near Bozeman Montana. There the mid-levels were set to discuss in front of the NSF and commercial service providers multiple scenarios for the continued development of the network. It was at this meeting that ANS formally unveiled its Plan for Commercial Services to all the mid- levels and the network at large. The plan was a refinement of the ideas discussed in the ANS Proposal of January 15th offering the mid-levels an opportunity to sign commercial service agreements with ANS. ANS very likely believed that in less than four months it would be blessed by the NSF as the provider of choice to operate the NSFnet backbone after the Cooperative agreement would expire in November 1992.

ANS Vice President Joel Maloff made a speech in which he showed how the market for competitive long distance providers underwent a shake out driven by the arrival of better financed companies. As Maloff put it "The lessons that I see for all of the members of FARNET are as follows:

Determine what you do well, and what you can do better than most others, and stress the building of those activities.

Do not try to hold the status quo against bigger, better funded, more able organizations. Consider outsourcing and alliances. Bend with the wind and move to serve your constituencies rather than remaining intransigent.

Diversify your markets and sources of supply so that you are not captive to any one source and therefore vulnerable to whims and sudden changes."⁷⁴

A summary of the explanation was included in a two page document called A Mid-level's Guide to the ANS Agreements and dated August 14, 1991.⁷⁵ The purpose of the agreements was to find a framework that would support "mixed research/education and commercial use of FARnet mid-levels and the ANS network. . . . . The implementation of this framework has been cast into a series of agreements. In a "normative case" a FARnet member would sign the first of them, the Connectivity Agreement and either the Gateway Attachment agreement, or the Cooperative Agreement or a variant."

By signing the Connectivity Agreement, the mid-level would permit commercial traffic from the ANS backbone to enter its network. It would also agree to participate in the National Infrastructure Pool and be eligible to receive funds there from. So far so good, but what ANS perhaps inadvertently ignored was that commercial sites attached to mid-levels were free to use the NSFnet backbone (the same physical circuits as the ANSnet backbone) if they pledged that such use was in support of research and education. Some find it hard to imagine that clearance for more obvious commercial use would prove that attractive to organizations attached under the "com" domain that were there in the first place primarily to communicate with R&E organizations in support of research and education. But others say that "even though a corporation is already attached as R&E, it doesn't mean that they are not attracted to the potential of not having to worry about the NSFnet acceptable use policy (AUP) and are willing to pay more for that privilege. Many corporations are deliberately limiting their use of the internet because of AUP concerns.⁷⁶

If a mid-level signed the Gateway Agreement it would keep control of its commercial clients for purposes of billing and support and would pay ANS fees based on estimates of the amount of commercial traffic that they would send across the ANSnet backbone. The fee structure was based on a variant of the fee structure first elaborated in the January 15 proposal discussed above. While the more detailed ANS Plan for Commercial Services showed a complicated formula by which these fees would be measured, it did not give information in enough detail so that a third party could come up with a dollar estimate figure of what it would cost a given mid-level to sign the agreement. A best guess would be about $2000 per commercial site per mid-level would go to ANS. While the commercial sites connected to some mid-levels only number about 25, other mid-levels have 100 or more commercial sites. This could become rather expensive considering that a mid-level could join the CIX for a flat $10,000 -- giving all its commercial sites unrestricted use of the CIX backbone. Unfortunately the drawback to this solution was that about 35% of all network sites were reachable only by transit of the ANS backbone. For some mid-levels another drawback was the cost of running a T-1 link from their location to the CIX interconnect point. And, in the year following the debut of the Agreements at least two mid-levels found it cheaper to sign with ANS rather than joining the CIX and running the connect line (in the case of one) or merely running the connect line (in the case of the second).

If a mid-level signed the Cooperative Agreement, the result would be "a closer and less hierarchical relationship" than would apply to the Gateway agreement. The mid-level would generally let ANS CO+RE market commercial connections directly to the mid-level. By in effect giving up a portion of its business to ANS, the mid-level would not have to worry about the expensive connectivity fees that came with the Gateway agreement.

Given the way that ANS had aggressively moved into the mid- level's territory over the past year, it is not difficult to guess that most mid-levels at the Montana FARnet meeting were not very comforted by what they saw of the implications of their commercial use of the ANS backbone.

As reported to the com-priv list by Robert Weber, the consultant who served as a planner facilitator at the FARnet meeting, a series of four scenarios for dealing the problem presented by the expiration of the backbone were discussed. It had become clear that if the NSF backed out of the picture, given the opposing views of network connectivity on the part of ANS and the CIX, instability and possibly a split of the network into two hostile camps lay ahead. This was not what FARnet members, especially the mid- levels wanted. Therefore according to Weber, "at the end of the second day, participants felt that predictability and stability were critical factors in any scenario going forward. The group [also] felt that the NSF should continue to play a lead role in any future interregional connectivity program. Two solutions were favored:

(1) extending the current contract with MERIT/ANS for some period of time, and (2) creating a new solicitation for interregional connectivity after November 1992, and then awarding a contract."⁷⁷

In other words the mid-levels wanted the NSF's continued involvement as court of last resort in dealing with ANS's expectations.

On November 22, 1991 they got almost precisely what Weber said they wanted as the National Science Board approved the request by the NSFnet Director to extend the MERIT contract by 12 to 18 months while the backbone was put through a re solicitation process.

A Flap Over ANS's First Commercial Customer

The next paragraphs explained what this would mean: "ANS commercial traffic will be restricted from interchange with federally sponsored networks which have not signed an ANS Connectivity Agreement or developed an alternative plan with ANS that will meet their requirements. This restriction on ANS commercial traffic also covers other Internet connected networks such as state, federal agency and international networks that do not wish to receive commercial traffic.

For the interchange to occur, federally sponsored mid-levels must have in place a signed connectivity agreement with ANS which has been designed to meet the requirements established by the NSF. Participation in the basic connectivity agreement between a mid- level and ANS is made without any charges to the mid-level. ANS connectivity agreements are currently in place for several mid- levels: BARRNet, MichNet, MidNet, OARNet, MRNet, NYSERNet and PREPnet; several others are in the process of final approval. As the list grows, it will be posted. Customers at campuses and research organizations attached to these networks will be able to access ANS connected commercial service providers directly via the Internet." The announcement went on to add that the filtering would take place by leaving paths to non compliant networks out of the routing table of the router placed at the premises of Dialog - ANS' first commercial customer which would be subject to the policy.

The statement that, "participation in the basic connectivity agreement between a mid-level and ANS is made without any charges to the mid-level," tended to be overlooked in the ensuing controversy. It is easy to see why. This was either a shift in or clarification of ANS policy, for the Mid-level's Guide to the ANS Agreements cited above stated that in a normative case mid-levels were expected to sign the connectivity agreement and either the Gateway Agreement or the Cooperative Agreement - both of which had significant financial implications for the mid-levels.

Apparently ANS was leaving the door open for mid-levels to sign only the Connectivity Agreement. With this very important nuance left unclear it was easy for readers of the announcement to assume that the route blocking tactics were being used by ANS as a squeeze play on those mid-levels who had not signed ANS agreements that would either place them under financial obligation to ANS or would limit their ability to market to commercial clients in the future.

Those institutions which were blocked from accessing Dialog had three choices: (1) convince their network provider to sign the ANS connectivity agreement, or (2) "Interconnect to a network service provider that participates in the ANS Connectivity Agreement," or (3) Establish an ANS CO+RE connection. This is a fee-based ANS attachment service for sites that choose to interconnect directly to the ANS network." The language of the preceding paragraph stated that the three points applied to "networks that have agreed to support the exchange of commercial traffic with ANS in one of the following ways:" As such the language is very confusing for mid- level networks were known to connect with the backbone and not with each other (2), and mid-level networks could hardly be considered to be "sites" (3). It is safe to assume that readers interpreted points 2 and 3 as invitations for institutions which were connected to nets blocking them from access to Dialog to either switch directly to the ANS backbone, or to a mid-level that had signed an agreement with ANS.

The next day (December 8, 1991) PSI's President Bill Schrader posted the NSF privatization agreements of September 10, 1990 and May 24, 1991 for the first time asserting that the Dialog route blocking should be viewed in the context of these documents which appeared to cede control of the network from the NSF to ANS.

On December 11, 1991 Mitch Kapor and Dave Farber (both Directors of the Electronic Frontier Foundation) authored an Open Letter which they posted to com-priv:

"Merit, NSF, and ANS have recently announced tighter restrictions on commercial traffic flowing across the backbone. Steve Wolff posed a clarification yesterday citing the reasons for this move.

This has set off a flurry of responses some of which question whether it is really meaningful to distinguish between 'research' and 'commercial' traffic, other of which begin to propose technically elaborate schemes to embody finer distinctions. Steve [Wolff] agrees the model of segregating traffic by network number is seriously broken, but that he is forced to use it by virtue of the obligations [placed] upon federal employees in the disbursement of federal support.

The evolution of a strong networking infrastructure is essential to the health of R&D in this nation. Competition has shown itself to be an effective vehicle for creating the best for consumers as well as providing jobs and trade. We believe that no further progress in networking infrastructure is possible without ensuring the creation of a level playing field.

The situation which will result from the latest mandates creates a market that is fundamentally unfair in that it will tilt the playing field too strongly toward one player -- ANS. Since the regional networks are physically connected to the backbone, they will be forced to execute an ANS agreement in order to receive any commercial traffic whatsoever. This creates an unfair marketing advantage for the ANS commercial business unit. As with any provider, ANS's commercial business unit is a welcome competitor in the market -- however, it must not receive an unfair competitive advantage owing to ANS's relationship with the NSF.

This advantage to ANS will have been accomplished solely by virtue of the exercise of intentional or unintentional NSF policy. No provider -- ALTERNET, ANS, CERFNet, PSI, etc. should be given such an advantage without public and open discussion and competition.

We are looking for ways to ameliorate the situation. We do not think that charges and counter charges however valid, are in the end, constructive. We are looking for concrete positive suggestions, either for a resolution process, or for elements of a 'level field' set of rules. We are anxious to receive such suggestions by private mail or public posting.

This was the first time that figures of the stature of Farber and Kapor had publicly stepped into the fray. Consequently on December 19, 1991 a New York Times article appeared under the byline of John Markoff. According to Markoff: "people involved in the planning for a national data network say it is essential to provide for fair competition which will lead rival companies to offer creative and entrepreneurial services in the hope of building market share. Without competition, they say the Government will have created a monopoly that has little incentive to innovate. . . . . A number of organizations are working informally to settle the dispute."⁷⁹

The spin that ANS put on the story emphasized the infrastructure pool as what - in the long run - should make everything seem worth while. Weis was quoted as saying that while funds in the infrastructure pool were "small - less than $25,000, and perhaps $250,000 in a year - it could amount to millions of dollars in just three to four years."⁸⁰

In response to the Communications Week article the author made a posting to the com-priv list which included the following statement:

"We read that a regional network that wants to send commercial traffic to other regional networks over the ANS network has to pay ANS a fee. This sounds to me like the Gateway agreement. If I read that correctly, such agreements could be quite costly to the mid-levels. However I have yet to see an example of exactly how much signing such an agreement would cost a mid-level."⁸¹ The next day the author received a message from a high level official of a mid-level network stating that he too had yet to see such an example. "We at XYZNet⁸² have tried to no avail to talk numbers with ANS, We find them to be arrogant, unwilling to listen, and generally impossible to deal with." After some discussion of the desires of the network's commercial customers, he concluded: "All of this puts us into a dilemma: if we don't sign with ANS for commercial connectivity how do we provide that service for those of our commercial members who want us to transport commercial traffic?

Later in the same day the author received a second message, this time from an employee of ABCNet⁸³ "ANS sent their salesperson over to feed us the party line. We discussed our needs and what we wanted from ANS. Seems they want to 'partner' with us but are unwilling to guarantee us a piece of our own pie. I suspect that ANS wants to use the regionals to penetrate the market and then displace the regionals in time. It is seductively easy to use ANS CO+RE instead of the CIX. ANS would keep track of the packets that travel between hosts in the COM. domain and bill the regional annually. The regional then passes the cost on to the commercial users. they also offer you a carrot: you are eligible to get back a share of the money that is collected based on your non commercial traffic. Sounds great, huh? But then I remember the meeting with the salesperson. They want to own the whole damned thing."

The author continued to ask questions about the ANS agreements. In conversations via private mail and on com-priv, it was pointed out that ANS would allow a network to a Connectivity Agreement without insisting that it also sign a Gateway or Cooperative Agreement. However the nuances that continued to be expressed by those speaking for ANS were such as to still leave the overall affect of what ANS policy would be on a given mid-level very unclear. Finally Dr James Bruce, Vice President for Information Services at MIT and a member of the National Academy of Sciences sent the following message to the com-priv list.

Gordon, I think you have captured quite eloquently in your questions the brokenness of the combit model and the ANS implementation of commercial and R&E traffic on the backbone. I learned long ago that you cannot build a strong lasting structure on a broken foundation. Yet that is what is being attempted.

Many of us who have responsibilities for the mid-levels are sitting on the side lines in these discussions because we understand that the present situation is broken. We see little opportunity to change this situation for the better given the proclivity of various individuals speaking now for rhetoric and their inability to listen. However we will continue to look for new solutions and I am sure that one that is "out of the box" will ultimately be found.⁸⁴

Meanwhile ANS pointed out for the first time that a network could sign only the Connectivity Agreement. However it added that while a network that signed this agreement could receive commercial traffic from the ANS backbone, it could not send commercial traffic across the backbone.⁸⁵

In mid January the issue of Dialog as ANS's first commercial customer began to boil up again when Geoff Goodfellow posted the following to com-priv:

I called up Dialog's Chief Counsel, Bob Simmons, this morning. I found out a few things that may be of interest:

1. When Dialog signed up with ANS, ANS didn't tell Dialog that their status as a commercial customer would prevent them from having 'full' Internet access.

2. Dialog published in their customer newsletter that they were now connected to the Internet and all you had to do was 'telnet dialog.com.'

3. Dialog found out they did not have 'full' Internet access only by customer complaints when users couldn't access the service with telnet dialog.com called in.

4. Before Christmas, Dialog requested that ANS change Dialog's status from Commercial to R&E. To this day ANS has not made the change from CO to RE."⁸⁶

The solution to the current dilemma has been for Dialog to adhere to the ANS research and education Acceptable Use Policy. This is an interim step in the correction of a larger acceptable use policy problem. and we have consulted with a variety of involved organizations, including Merit and NSF, prior to coming up with this answer. With the implementation of routing updates late on Thursday January 23, research and education acceptable users of the Internet will be able to access Dialog on the following day.⁸⁷

No matter how one looks at the outcome, the Dialog case strains one's credulity. Was communication within ANS so bad that Maloff was not informed that Dialog, as a first commercial client, would become the test case for implementing routing filters? Or ANS was so certain that its control could or would not be questioned, that it was blind as to how its actions would be perceived by its first commercial customer and by the network community? Moreover what is even more remarkable is that while it has signed up Orbit and BRS relatively recently, whether these clients are commercial or R&E adherent is generally unknown.⁸⁸ Because it is hard to understand how they would agree to terms less advantageous than Dialog received, one suspects that they are R&E. The author is unaware that they have been blocked from networks that have not signed the ANS agreements.

Maintaining an Integrated Internet: ANS and the CIX

The ANS CIX interconnect agreement was announced in late May. It appeared to paper over diametrically opposed stances on the part of the two parties. Many network users seemed to think that it was a complete interconnection that would open the flow of traffic freely in both directions. In reality this was not the case. Because ANS is not a member of the CIX, its retail customers such as Chevron, Union Carbide, and Abbot Labs are not permitted to send their data through the CIX. "ANS is acting like a telephone company where it brings its wholesale customers traffic to the CIX, if and only if that wholesale customer is also a CIX member independently. Think of ANS as an alternative to renting a circuit from NEARnet to Santa Clara, and nothing more."⁹⁰ The statement about renting a circuit refers to the case of NEARnet, a CIX member situated primarily in the Boston metropolitan area that was able to negotiate, what apparently is a modified gateway agreement with ANS, to allow its commercial clients to use the ANS backbone to transit to the CIX interconnect point in California. Whether ANS would pull out in June of 1993, if it didn't get its way with settlements in the meantime was anyone's guess.

In the meantime ANS continued work on its "Mechanism for Capital Infusion into the Backbone Via Mid level Network Fees and Settlements for Commercial Traffic" as Appendix C of its January 1991 "Proposal" to the NSF had been called. With mid-level networks appearing to be able to get something (eligibility for the Infrastructure Pool) for nothing by signing the Connectivity Agreements, ANS was able to gradually increase the number of mid-levels that were willing to take at least the first step in forming commercial relationships with it.