A trader presses a key and is already speaking. No dial tone, no ringing, no hello. Chapter One

The channel has been open since seven this morning and nobody joined it, because a system that announced who was listening would be useless.

Behind that key sits forty years of accumulated engineering, most of it undocumented, some of it accidental, and almost none of it written down anywhere a modern team would find it. The trading floor is the last place in finance still running a device that feels like 1987. This is the story of what happened when someone tried to replace it.

Open Line follows a rebuild of that system on WebRTC and Kubernetes, and asks one question of every feature, old and new.

The question the book keeps asking

Is this the invariant, or is this the implementation?

The answers are frequently uncomfortable. Most of what looks like legacy turns out to encode an answer somebody would otherwise have to rediscover. Some of what looks essential turns out to be a period detail nobody revisited.

A latency budget that is not about speed

It turns out to be about certainty. The number traders defend is not the one an engineer would have optimised.

A recording that satisfies the regulation

And cannot answer the question a regulator actually asks. Compliant and useless are not mutually exclusive.

A migration whose unit is not a desk

It is the smallest group of people who have to behave as if they were one floor. Get the unit wrong and nothing else matters.

A vendor deliberately unable to prove who its customers are

Each firm keeps its own authority. The vendor is a relying party and nothing more, on purpose.

A negotiation that ran into its third year

Over a private call, an employee representative body, and whether a person may ever switch the recording off.

A second telephone on the desk

Which is still the only answer anybody has found. The book declines to pretend otherwise.

Background, for anyone new to the floor

What is a trading turret, and what does WebRTC change?

Trader voice is the industry's name for this whole category: the specialised voice systems a trading floor runs instead of ordinary telephony, built for people who talk to the same counterparties every day and cannot wait for a call to connect. Its centrepiece is the turret.

A trading turret, also called a dealerboard, is the dedicated console on a trading desk: a bank of sixty or more physical keys, each one a standing, pre-provisioned line to a specific counterparty or channel. It is not a telephone with speed dials. Pressing a key closes a circuit that is already open, so there is no dial tone, no ringing and no greeting. The trader is simply speaking.

That behaviour comes from the private wire underneath it, a permanent leased connection negotiated once at provisioning time, sometimes years before anybody presses the key. It also explains the other thing floors run: the hoot, an always-open broadcast channel nobody answers and nobody joins, because a system that announced who was listening would be useless.

WebRTC changes the endpoint. The turret becomes a browser tab or an application on ordinary hardware, media travels as SRTP over a negotiated peer connection rather than down a dedicated wire, and the session is established per press instead of once per contract. Put that on Kubernetes and the media plane becomes something you scale, drain and fail over like any other workload, except that it is stateful and somebody is talking through it.

What does not change is the interesting part. The trader still has to know, in the instant their finger lands, that they are live. The recording still has to satisfy a regulator. The counterparty still has to be provably who they claim to be. Open Line is the story of taking a turret apart on modern infrastructure and finding out which of its forty-year-old behaviours were period detail and which were load-bearing.

Thirty-five terms a trading floor never defines, or read Chapter One free.

Drawn, not just described

Thirty-three diagrams, and a companion Visual Atlas

Every argument that can be drawn is drawn. Each figure states what to take away from it, so the atlas works on its own.

Three clocks diagram showing where a call's time is actually spent. — Where the time actually goes, and which part of it a trader can feel.

Diagram contrasting a registered endpoint with a connected one. — Registered is not connected. The distinction costs more than it looks like it should.

Diagram showing media placed beside the orchestration plane rather than inside it. — The media plane has one job, and it belongs beside the control plane, not inside it.

Diagram showing three classification bins where teams usually assume two. — Three bins, not two. The middle one is where the arguments happen.

Advance praise

What readers in the industry are saying

“The story of how Steve and Amar developed a new trading voice system by replacing wires and exchanges with software and servers.

“A must read for anyone involved in trading floor voice systems and trying to replace the old with the new.”

M.L., fifteen years supporting trading voice systems for Tokyo financials

Contents

Twenty pieces, one argument

The first half takes the old machine apart. The second half rebuilds it and finds out which parts refuse to move.

1 The Button That Cannot Wait Three clocks under one key press, and which one a trader can feel Free

2The Call Nobody AnswersThe always-open channel, and why a hoot has no ringing

3Fifty Buttons and No MenuSixty-two keys, recognition over recall, and the key sheet

4The Wire That Became a RelationshipWhen a private wire stops being a circuit and becomes a counterparty

5The Sounds That SurvivedDial tone, comfort noise and sidetone as load-bearing engineering

6Is the Turret the Product?What a bank is actually buying when it buys the hardware

7Cloud Native, Not Cloud CarelessStatefulSets, disruption budgets, and draining a node with a live call on it

8Identity Is the New Line CardFederated PKI, and a vendor that cannot prove who its own customers are

9The Media Plane Has One JobMixing and forwarding beside the control plane, never inside it

10Recording Is Part of the CallRecording that satisfies the regulation and cannot answer the question

11The Floor That Runs BothMigrating a floor when the unit is not a desk

12Two Data Centres Are Not ResilienceFailure domains, and degraded as a state you can actually see

13APIs for Things That Used to Be ButtonsReads, writes and acts, and which of them a turret may safely expose

14The Desk Becomes ProgrammableWhat happens when the desk starts taking instructions

15The Conversation Can Understand ItselfTranscription, analytics, and the argument about who is listening

16The Bank Can Change Vendors Without Changing RealityPortability, and the moat that was never technical

17What Would Not MoveThe ledger: what stayed, what escaped, and what was never built

Epilogue: The Third ChairEvery room in the book had two sides, and one more chair

AfterwordWhat the book never managed to write, listed rather than omitted

GlossaryThirty-five terms a trading floor uses and never defines

What this book is not

It is not a success story

The platform runs. The floors that adopt it keep it. The incumbents are never displaced. What spreads is the argument.

Written by two people who have built and sold this technology, and narrated by an outside observer whose independence the book examines rather than assumes, Open Line is a design book that refuses the usual comforts. It states what it does not know. It quotes the numbers you can falsify and withholds the ones that would flatter it. Its final chapter is a list of what it never managed to write.

For architects, engineering leaders, and the people who own the budget and the risk.

The authors

Both sides of the contract

Steve Phillips

Co-author

Has spent his career in trading-floor communications, including as CEO of Infinet Financial Systems and running Asia-Pacific for a turret vendor out of Tokyo. He has been on both sides of the contracts described in this book.

Amar Akshat