# What is TheHUB

TheHUB is neutral colocation infrastructure for latency-sensitive markets. It leases space and power in a colocation data center, currently Equinix TY15 in Tokyo, and builds its cage there. Inside its cage, TheHUB connects trading venues, trading firms, validators (blockchain nodes with a consensus role), and the computing providers that host other firms' servers, over private fiber, with no public internet between them. It does not run a trading venue, match orders, hold positions, or trade, and it is not tied to one asset class.

To connect, start at [Onboarding](<https://docs.jointhehub.com/onboarding/index.md>).

## The problem TheHUB addresses

Where firms and venues colocate in one data center, position decides speed. TheHUB removes each of these effects for its participants:

| Where position decides speed | On TheHUB |
| --- | --- |
| **Nearest gets the shortest cable.** An adjacent cabinet has a shorter fiber run to a venue than a distant one. | **The same target fiber length** from every participant to a given spine, one of TheHUB's central switches, whatever the cabinet's position (see [Equal path lengths](<#equal-path-lengths>)). |
| **Same-switch turns.** Two systems on the same switch exchange traffic there, skipping the hops a more distant pair crosses. | **Every flow crosses a spine switch,** even between two participants on the same leaf (a switch where participants plug in); see [Forced spine transit](<#forced-spine-transit>). |

This applies among TheHUB's own participants; nothing here is claimed about paths outside TheHUB.

## How TheHUB works

[![Diagram of TheHUB's site, described in the table below.](<https://docs.jointhehub.com/assets/images/thehub-overview-wide-light.svg#only-light>)](<https://docs.jointhehub.com/assets/images/thehub-overview-wide-light.svg>) [![Diagram of TheHUB's site, described in the table below.](<https://docs.jointhehub.com/assets/images/thehub-overview-wide-dark.svg#only-dark>)](<https://docs.jointhehub.com/assets/images/thehub-overview-wide-dark.svg>)

[![Diagram of TheHUB's site, described in the table below.](<https://docs.jointhehub.com/assets/images/thehub-overview-stacked-light.svg#only-light>)](<https://docs.jointhehub.com/assets/images/thehub-overview-stacked-light.svg>) [![Diagram of TheHUB's site, described in the table below.](<https://docs.jointhehub.com/assets/images/thehub-overview-stacked-dark.svg#only-dark>)](<https://docs.jointhehub.com/assets/images/thehub-overview-stacked-dark.svg>)

*TheHUB's site at Equinix TY15 in Tokyo. Green marks the equalized path: every participant's path to a given spine, on either route, built to the same target fiber length (select the diagram to enlarge it).*

| Element | What it is |
| --- | --- |
| Network cabinet | The cabinet in TheHUB's cage holding its network equipment, including its [DoubleZero device (DZD)](<#the-doublezero-device>) |
| Spines | Spine switches joining the leaves in a leaf-spine fabric. Market data and order entry run on [dedicated physical and data paths](<#dedicated-paths>), with failover |
| Cage leaves | Two leaf switches for equipment in TheHUB's cage, patched with no cross-connect. Both carry market data. Their uplink runs through the uplink spool, fiber that makes up the rest of the path length |
| Equipment in the cage | For example, own servers, a venue's matching engine, or a validator |
| External participant | A participant with equipment in its own space in the same building |
| Demarcation | Where the participant's own network hands off. The equalized path runs from here to a given spine; the participant's side is outside it |
| Fiber coil | The compensating spool, fitted at the demarcation to make up the length |
| Cross-connect | Fiber ordered from the data-center operator, from the participant's space to the border leaf, inside the equalized path |
| Border leaf | Where participants arriving over a cross-connect land. There is one border leaf, and it carries no in-hub market data |
| DoubleZero device (DZD) | TheHUB's node on DoubleZero, a private fiber network that carries market data between regions |
| FPGA: timestamps and multicast | A field-programmable gate array (FPGA) appliance in TheHUB's cage, hardware that runs its logic in silicon at wire speed |
| GPS clock | TheHUB's time reference, a Global Positioning System (GPS) clock |
| AWS gateway | Amazon Web Services (AWS), reached from the border leaf over a direct link: a cross-connect to AWS Direct Connect that TheHUB provides on request. With no compensating spool, it sits outside the equalized path |

| Term | What it means on TheHUB |
| --- | --- |
| **Participant** | Any party whose equipment connects to TheHUB |
| **Tenant** | What TheHUB provisions, with its own routing table, role, and address block. A participant is one or more tenants |
| **Computing provider** | A provider that hosts other firms' servers: a bare-metal host or an on-premises cloud |
| **Downstream customer** | A firm whose servers a computing provider hosts; each is a tenant |

### Forced spine transit

- Leaves attach to the spines only; there is no leaf-to-leaf link.
- Leaves hold no default route and never pass routes between tenants' routing tables, so two tenants reach each other only through a spine, even when both sit on the same leaf.
- Every unicast flow between two tenants therefore crosses exactly three switch pipelines (ingress leaf, spine, egress leaf).

**Naive fabric**

**TheHUB**

*In the naive fabric (a counterexample, not TheHUB infrastructure), a same-leaf pair's traffic turns at the leaf; on TheHUB it goes up through a spine and back out of the same leaf (heavy line).*

### Dedicated paths

Market data and order entry each have a dedicated path through TheHUB's core (its spine switches, which every flow crosses), on separate physical links and switching, so neither queues behind the other. Both are redundant in the core: a single core switch or link failure does not take either one down. During a failover, the two may share hardware until the fault is repaired. The redundancy is in the core only: the border leaf, and a participant's single port on one cage leaf, remain single points of failure.

Switch time differs between the two paths, and fiber equalization (see [Equal path lengths](<#equal-path-lengths>)) holds within each, not across them.

### Equal path lengths

TheHUB builds every participant's path, from its demarcation to a given spine, to one **common fiber length**, on either route.

| Route | How TheHUB brings it to the common fiber length |
| --- | --- |
| **The cross-connect route**, from a participant's own space in the building | TheHUB measures the cross-connect run and fits a **compensating spool**, a coil of fiber cut to make up the length, at the participant's demarcation |
| **The in-cage route**, from a cabinet in TheHUB's cage | TheHUB runs equal-length fiber from its network cabinet to the patch panels at the top of every cabinet in the cage, so a cabinet's position does not change its path length. A spool of fiber on the cage leaf's uplink, the **uplink spool**, makes up the rest of the length |

Equal fiber length gives every participant the same fiber delay, D, to a given spine, so any two participants are 2D apart one way in fiber (participant to spine to participant).

TheHUB cuts each equalizing fiber to within plus or minus 10 mm of its target length. That is about 49 picoseconds of one-way propagation at about 4.9 nanoseconds per meter of standard single-mode fiber. This is a worst-case bound from the cut tolerance, not a measurement. The tolerance bounds the cut only: the measured cross-connect run, connectors, patch cords, and the participant's own side of the demarcation are outside it.

> [!NOTE]
>
> **Uncongested fabric**
>
> Equalization applies to an uncongested fabric: a queue adds delay that no spool removes.

### Isolation and role-based reachability

Each tenant terminates into its own **Virtual Routing and Forwarding (VRF)** instance: an independent routing table on the shared switches. Routes pass between VRFs only at a spine, and only as policy allows.

Policy follows each tenant's **role**, agreed at the [kickoff](<https://docs.jointhehub.com/onboarding/index.md#the-three-step-sequence>). TheHUB splits tenant address space into an exchange aggregate and a trader aggregate, each a summarized address block, and advertises each tenant the opposite role's aggregate and no same-role route.

The role decides which other tenants a tenant can reach. TheHUB recognizes two:

| Role | For | Reaches | Does not reach |
| --- | --- | --- | --- |
| Exchange role | A trading venue | Every tenant in the trader role | Any other tenant in the exchange role |
| Trader role | A trading firm | Every tenant in the exchange role | Any other tenant in the trader role |

- **A firm that is both a venue and a trading firm** is provisioned as two tenants.
- **Changing a tenant's role later means renumbering it.**
- **Other participant types,** such as a validator (a blockchain node with a consensus role) or an oracle publisher (a participant that publishes price data onto a blockchain), also take one of the two roles, chosen by what the tenant must reach. The role is agreed at the kickoff.

### Market data

TheHUB distributes market data inside TheHUB as Internet Protocol (IP) multicast, one stream delivered to many receivers, to equipment in TheHUB's cage.

1. A venue in TheHUB's cage sends its feed to TheHUB as unicast User Datagram Protocol (UDP) on its **outbound market-data port** (see [Ports for a venue](<https://docs.jointhehub.com/onboarding/venue-or-trading-firm/index.md#ports-for-a-venue>)).
2. The FPGA appliance turns the feed into IP multicast.
3. The stream runs on the market-data path, through the cage leaves, to the subscribers in the cage.

Placement decides whether you receive in-hub market data; your path length is equalized on either route.

| Who | Receives in-hub market data |
| --- | --- |
| A participant's own equipment, patched to a cage leaf | Yes. It joins a feed with Internet Group Management Protocol (IGMP), handled in the FPGA appliance rather than on the switches |
| A computing provider's downstream customer, hosted in TheHUB's cage | Yes, where the provider's first-hop switch runs the multicast control plane and carries the join to TheHUB (see [Onboarding a computing provider](<https://docs.jointhehub.com/onboarding/computing-provider/index.md#receive-in-hub-market-data>)) |
| A participant in its own space, over a cross-connect to the border leaf | No |

Multicast is the default. A participant that cannot ingest it can take a **dedicated market-data port** instead.

Market data from other regions reaches TheHUB over DoubleZero (see [The DoubleZero device](<#the-doublezero-device>)).

### The DoubleZero device

TheHUB operates the DZD on DoubleZero as a contributor, an operator that adds its own device to the network.

- **Placement:** on the market-data path.
- **TheHUB's infrastructure, not a participant:** the DZD has no VRF and no role.
- **Advertised to every tenant:** its user-tunnel endpoints, where tenants' tunnels terminate, so a tenant that uses DoubleZero (which is optional) reaches the DZD across TheHUB's fabric.
- **Private.** The user-tunnel endpoints are not reachable from the internet, only across TheHUB's network.
- **Built from the host:** the DoubleZero client software on the participant's host builds the tunnel to one of those endpoints (see [Set up the DoubleZero client](<https://docs.jointhehub.com/market-data/doublezero-client/index.md>)).

*Through the DZD, a venue publishes its market data onto DoubleZero as IP multicast, and trading firms subscribe to it.*

TheHUB's DZD copies the venue's stream into each subscriber's tunnel. The equalization on this page covers each participant's path to a given spine, not the DZD's own link into TheHUB's fabric.

## What you connect to

Speeds are in gigabits per second (G). TheHUB sends you an **allocation table** with the values it assigns you.

| Parameter | Value |
| --- | --- |
| Physical handoff | A standard data-center handoff, one of: a cross-connect from your own space, a patch inside TheHUB's cage, or a connection through a computing provider that is already connected to TheHUB |
| Attachment | A **dedicated port**: a routed physical port for one tenant alone, used by every venue or trading firm and by a computing provider's downstream customers on bare metal. A venue in TheHUB's cage takes at least two, as separate ports: an outbound market-data port and an order-entry port. A **trunk** is one port carrying a computing provider's downstream customers on cloud (virtualized) servers, each tenant on its own routed sub-interface |
| Dedicated port speed | 10 G standard, 25 G on request |
| Trunk port speed | 100 G |
| Leaf-to-spine links | 100 G |
| Peering link | A /31 from Request for Comments (RFC) 6598 shared address space (100.64.0.0/10), in your allocation table. Never advertised |
| Peer Autonomous System Number (ASN) | TheHUB's single ASN, in your allocation table, the same wherever you attach. You peer over external Border Gateway Protocol (eBGP) and see one TheHUB ASN in the AS path, and none of TheHUB's internal structure |
| Your address block | From TheHUB's public Internet Protocol version 4 (IPv4) space, inside the aggregate for your role, in your allocation table |
| Routing policy between tenants | TheHUB accepts only the address block registered for you and, by a deny-own-community rule, never advertises you your own routes. You receive the opposite role's aggregate and the DZD's user-tunnel endpoints, and nothing else: no same-role route and no default route. This site does not describe routing for internet or AWS access: if your order includes either, ask your TheHUB contact |
| Path Maximum Transmission Unit (MTU) | 1500 bytes, the standard MTU. Do not configure jumbo frames toward TheHUB |

## See also

- [Onboarding](<https://docs.jointhehub.com/onboarding/index.md>): which onboarding guide to follow.
- [TheHUB: public brief](<https://jointhehub.com/thehub.md>): the one-file description of TheHUB.

Last reviewed 2026-11-20.
