Mark Nichols

The Digital Island Litmus Test and Carrier Architecture Record, 1996

More than forty incumbent carriers and ISPs. No competing offer to Digital Island’s service proposition.

In 1996, there were forty-three major carriers and ISPs including AT&T, MCI, Sprint, WorldCom, Global One, Pacific Bell, Southwestern Bell, Bell Atlantic, NYNEX, Ameritech, US West, BellSouth, British Telecom, Cable & Wireless, France Télécom, ImagiNET, Deutsche Telekom, KPN, HiNet, Telia, Telefónica, Telstra, NTT, Japan Telecom, KDD, IDC, ITJ, Telekom Malaysia, SingTel, Korea Telecom, SK Telecom, Embratel, TelMex, Rostelecom, Bezeq, NetVision, EasyLink, UUNET, BBN, GTE, Exodus, AboveNet, PSInet, but none had the product proposed and provisioned by Digital Island:

The first Tier-0 global network. Operating under its own AS6553 and deploying private IPLC CBR circuits, Digital Island delivered a network architecture I initiated and branded as Merchant Transport.

By fundamentally bypassing the erratic, hot-potato public BGP backbone routing and eliminating localized peering point congestion, this architecture established the industry’s first deterministic layer for global transaction states. It directly remedied cross-border packet jitter, packet drop, and multi-second TCP window stalls that previously timed out early cryptographic transactions, ensuring a seamless transmission pathway between disparate sovereign infrastructure platforms.

For the first time, this fabric delivered enforceable, sub-300ms roundtrip SSL session completion QoS across the major internet markets, making cross-border commerce reliably repeatable at operational scale to ≈99% of all internet-accessible users.

The Governing Question: Why the incumbent product stopped at the border, and what Digital Island built past it

When Digital Island executed its first customer contract in November 1996, the global telecommunications market was not empty. It was saturated. AT&T, MCI, Sprint, WorldCom, British Telecom, France Télécom, Deutsche Telekom, NTT, KDD, Cable & Wireless, Telefónica, Telstra, SingTel, and their peers were all operating, all better capitalized, all with decades of plant and carrier relationships behind them. They were selling international connectivity to multinational enterprises, and they were selling a great deal of it. The question this record answers is not why anyone chose a three-person startup over them. It is narrower and more useful. What, structurally, did the incumbent product stop short of, and why did its architecture and its revenue model both make stopping there the rational choice?

The answer is structural, and it rests on four architecture facts of the 1996 market.

Part II: Six Deployment Records

Each entry states what the customer already had, what the incumbent product did not reach, what Digital Island deployed, and the dated artifact that proves it. No rhetorical questions. The facts carry it or they don’t.

1. Cisco Systems, November 1996

What Cisco already had. Everything relevant. The routers and switches running the Internet were Cisco’s own. It had engineering staff able to evaluate any vendor claim on the merits, unlimited capital, and standing relationships with every major carrier on earth.

What the incumbent product did not reach. Distribution of Cisco IOS images to engineers worldwide. A full feature-set BGP-4 kernel for a 7500-series router ran roughly 16MB. On oversubscribed Frame Relay, that transfer crossed multiple carrier boundaries, each with its own discard behavior under congestion. A packet dropped at 14MB timed out the TCP session, and the transfer restarted at zero. No carrier could commit to the end-to-end behavior the transfer required, because no carrier controlled the end-to-end path.

What Digital Island deployed. IPLC circuits terminated into Tier-1 backbone-facing ports under AS6553, removing DE-bit discard exposure from the path. Contractually defined service behavior across the international route.

The record. Remote Data Services Agreement, effective November 1, 1996, executed November 7 to 8, 1996, negotiated and signed for Digital Island by Mark Nichols. Contract value approximately $300,000. The agreement specifies a globally mirrored, load-balanced, fully redundant host with failover; a network segment reserved for Cisco; secure transaction services including encryption; a Service Level Attachment; and one-hour response.

2. Stanford University, January 1997

What Stanford already had. One of the strongest network environments in academia. BARRNet lineage, research backbone participation, and engineering staff drawn from the people who built the Internet. This record does not claim Stanford was an ordinary campus.

What the incumbent product did not reach. International delivery of HighWire Press journal objects, PDFs running 50MB to 100MB and larger. At intercontinental distance over oversubscribed transit, a single discard late in a large fixed object corrupted the transfer and forced a restart. The economics that followed were physical: print runs, warehousing, international shipping, and invoicing cycles measured in weeks.

What Digital Island deployed. Six cabinets leased on Stanford premises in the first week of January 1997, establishing Digital Island’s first Northern California Point of Presence on IPLC and ATM-switched architecture. From Q1 1997, Digital Island served as an upstream provider into the Stanford environment alongside Stanford’s existing research connectivity. This was also the operational correction to the Honolulu Frame Relay configuration, which December 1996 traceroute testing had shown tromboning Hawaii-to-Asia traffic through the mainland. Stanford became the second customer that year.

The record. Premises lease executed January 1997 by Mark Nichols; subsequent services agreement executed by Darren Hong. Two separate instruments.

On Google, stated precisely. Page’s crawler began running in March 1996, and google.stanford.edu was producing results on Stanford’s inherited infrastructure before Digital Island had a circuit on campus. This record does not claim otherwise. What the record does show is that from Q1 1997 forward, Digital Island cabinets, circuits, and upstream capacity operated on Stanford premises during the period when the crawl scaled from prototype to full index. The deployment sits inside that window alongside Stanford’s own backbone upgrades of the same era. Both are named. Neither is erased.

3. Visa International, Q2 1997, and MasterCard, 1999

What they already had. Brand, banks, capital, and relationships with every carrier under discussion.

What the incumbent product did not reach. Dependable cross-border completion of browser-based SSL sessions. One precision note, stated up front because this record is written for people who know the difference: card authorization, clearing, and settlement continued to run on the card networks’ own private systems. That was never the gap. The gap was the consumer-facing layer, and specifically the session. A discard event mid-handshake left a transaction neither completed nor cleanly failed. That ambiguity produced capital risk and manual reconciliation downstream, and it was a property of the transport path, not of the payment system.

What Digital Island deployed. End-to-end IPLC transport under AS6553 carrying browser access to these institutions’ online services over a controlled path rather than best-effort public transit. Sessions completed or failed inside a defined window instead of hanging.

The record. Visa signed in Q2 1997 as the third enterprise customer, within six months of the September 1996 Merchant Transport pivot. MasterCard press release, March 8, 1999, announcing Digital Island for virtual server distribution in key regions, localized hosting, application delivery, and ATM locator services.

4. E*TRADE and Charles Schwab Online, 1998

What they already had. Brokerage software, financial credibility, customers, capital, and carrier relationships.

What the incumbent product did not reach. Session persistence under international load, with routing stability and enforceable performance. Trading is less forgiving than publishing. A failed page load is an annoyance. A stalled order path is a filled or missed position and a compliance event. Public transit in 1998 could display information across borders. It could not commit to the behavior a regulated order path requires, for the same reason it could not commit for Cisco: no signer controlled the whole route.

What Digital Island deployed. The same fabric, carrying two regimes at once. Visa and MasterCard payment traffic and E*TRADE and Schwab order-execution traffic ran over one controlled path under AS6553, with backbone-facing terminations across six continents.

The record. E*TRADE strategic investment and board involvement following its adoption of Digital Island services.

5. Sun Microsystems and Inktomi, December 1999

What they already had. Sun built the servers the commercial Internet ran on. Inktomi built the caching and search software behind a large share of its traffic. Between them, the hardware and the software layer of content delivery.

What the incumbent product did not reach. Distribution of the payload. A Solaris recommended patch cluster ran 50MB to 150MB, and an engineer in São Paulo or Seoul pulling one across oversubscribed transit lost the session late and restarted at zero — Cisco’s IOS failure mode at larger payload and higher frequency. Sun could build the server and write the patch. It had no path that delivered either reliably at intercontinental distance, and neither did the carriers selling it transit.

What Digital Island deployed. Sun Netra servers onto the six-continent IPLC fabric, serving the payload from inside each market rather than hauling it across the public Internet on request.

The record. EE Times, December 8, 1999: approximately $25 million in combined minority equity from Sun Microsystems and Inktomi; up to 5,000 Sun Netra servers for global deployment; up to $150 million in network expansion targeting 350 additional metropolitan areas; 7 global data centers with regional expansion underway.

6. Microsoft, Intel, and Compaq, June 2000

What they already had. The operating system layer, the semiconductor layer, and the server layer of the Internet economy, and between them more engineering talent and capital than any comparable combination.

What the incumbent product did not reach. A unified delivery fabric for broadcast-scale streaming. Broadcast had towers, cable, and satellite. Carriers had transit. Hosting companies had rack space and power. None of those, separately or together, constituted a global Internet media distribution platform with performance accountability across markets.

What Digital Island deployed. Extension of the six-continent IPLC fabric into streaming infrastructure, with server deployment and capacity engineered against a stated target of up to 7.5 million simultaneous global viewers.

The record. $45 million private equity investment and more than 8,000 dedicated servers, June 2000. CBS MarketWatch, June 20, 2000, reporting Digital Island’s plan to build what it described as the world’s largest streaming media architecture with Compaq, Intel, and Microsoft.

The Operator Exclusion Table

Anchored to the 1996-era market. “Disqualifier” states the structural reason no competing product existed; “anchor” cites the documented proof.

Several operators on this list appear twice in the record: once as an entity with no competing product, once as the vendor that sold Digital Island the component. That is not a contradiction. Every physical element of the Tier-0 fabric was purchased from the carriers whose own products stopped at the border. They sold the parts. None assembled them.

Post-1996 entrants, noted for completeness and excluded from the count. Concert, in its 1996-era form the BT/MCI joint venture and from 1998 the AT&T/BT venture, was built specifically to serve this market, and in both incarnations delivered a carrier-managed product inside the carriers’ service envelope rather than autonomous customer routing under an independent AS. Combining incumbents reproduced the architecture; it did not replace it.

Independent Verification Layer

AS6553’s routing adjacencies are preserved in third-party archives: the University of Oregon Route Views project (snapshots from 1997) and CAIDA’s AS-relationship datasets (monthly from January 1998). Where a Digital Island backbone termination propagated to those vantage points, the adjacency is date-stamped, machine-readable, and recorded by an academic project with no stake in this history. One methodological note: presence in these archives is affirmative confirmation of a backbone termination; absence is not disproof, since private peerings did not always propagate to public collectors. Primary contracts remain the anchor evidence throughout.

The Record in Summary

Two operators on this list documented the exclusion in writing. Telefónica, with operations across Spain and Latin America, did not build a competing product; it signed a reseller agreement and sold Digital Island’s network to its own customers in April 1999. Cable & Wireless, with more global physical plant than Digital Island ever built, did not build a competing product; it paid approximately $340 million to acquire Digital Island in May 2001.

Forty-three operators. Six customer verticals. Roughly $70 million in strategic equity from Sun, Inktomi, Microsoft, Intel, and Compaq. 881 customers in under four years. There is no stronger form of carrier-exclusion proof than a transaction.

Scope and Method

This record claims that no operator listed offered, as a commercial product in the 1996-era market, an autonomous multi-continent IPLC fabric with enforceable end-to-end QoS under a single AS, and that the documented customer and investor decisions above are best explained by that absence. It does not claim these operators lacked engineering capability, and it does not address the post-1998 CDN era except where transactions speak for themselves. Corrections supported by documentation are welcomed and will be incorporated with attribution: mark@marknichols.com.

The Telefónica Reseller Agreement and the Cable & Wireless Acquisition of Digital Island: The Two Most Decisive Carrier Exclusion Proofs

There is no stronger carrier exclusion proof than a transaction.

Figure: Digital Island announces reseller agreement with Telefónica Data in Spain, April 26, 1999. Telefónica operated Spain’s national network and was expanding across Latin America through privatization acquisitions. It did not build a competing product. It resold Digital Island’s.

 

Figure: CNET News, May 14, 2001. Cable & Wireless acquires Digital Island for approximately $340 million. Cable & Wireless held submarine cable assets across more markets than Digital Island ever built. It did not build a competing product. It bought one.