11The European Map

On 6 June 2022 Euronext finished moving its Core Data Centre, and the colocation that went with it, from Basildon, east of London, to Aruba’s IT3 data centre near Bergamo, in northern Italy. The building it left was ICE’s; the building it moved to handles, in the words of Euronext’s chief executive, a quarter of European trading volumes. From Slough, where Euronext had put one of its London points of presence, light in a vacuum had needed $258\,\text{µ}\mathrm{s}$ to reach Basildon; it now needs $3.31\,\mathrm{m}\mathrm{s}$ to reach Bergamo. The firms that trade Euronext’s markets from London started that week four and a half milliseconds of straight glass further away, and two years later Euronext itself was selling them a microwave link to close part of the gap.

Europe has no single New Jersey. Its trading sites are spread over London, Frankfurt, Zurich, northern Italy, Stockholm and Paris, and the routes between them (London to Frankfurt above all) are the continent’s latency races. This chapter builds the European half of the site table with the same rules as chapter 10, adds the registry that says which venue matches in which building on which date, and measures what the migration did.

11.1 The London-area sites

A venue can be in a carrier-neutral building or in its own; what matters to a firm is that it can reach the venue’s systems, and its own carriers, from its cabinet. Euronext describes IT3 as “carrier-neutral by design”, with four carrier entry points (north, south, east and west) and dark fibre to the Milan Internet eXchange; Deutsche Börse’s co-location in Frankfurt is in a building where the housing contracts are between the firm and the operator, Equinix, and the exchange is “one of many customers”.

London has three sites for this chapter. Equinix’s LD4, on the Slough Trading Estate west of London, is where Euronext put one of its two London points of presence and where its microwave service to Bergamo starts; it is also one end of the London–Frankfurt routes below. The London Stock Exchange moved its primary data centre from a site at Liverpool Street, in the City, to Telehouse’s Docklands campus: planned for October 2022, the cutover of Millennium Trading, Turquoise and the Group Ticker Plant was moved to 18 February 2023. ICE’s European Liquidity Centre, in Basildon, Essex, runs ICE Futures Europe, and until June 2022 it also hosted Euronext.

The three London sites are close: 44 kilometres from Slough to Docklands, 34 from Docklands to Basildon, 77 from Slough to Basildon, that is 147, 114 and $258\,\text{µ}\mathrm{s}$ of light in vacuum. Inside such a triangle a firm can serve all three venues from one building, with a few hundred microseconds of glass to each; the migration took one corner of it to Italy.

11.2 Frankfurt and Zurich

Deutsche Börse’s page is precise: the Equinix FR2 data centre in Frankfurt is the co-location data centre “in which the primary Eurex and Xetra back-ends are located”. Its co-location connections are 10 gigabits, with transaction and market data on separate connections so that they do not queue behind each other (chapter 1), priced by the month: 6 000 EUR for one Xetra market-data or order-entry connection, 8 400 for two market-data products together, and 400 EUR for its PTP time service. Chapter 9’s American fee schedules are filed with the SEC; the European ones are published as price lists, and they are just as volatile.

SIX runs its co-location in Equinix ZH4 in Zurich and publishes an average latency of $10.6\,\text{µ}\mathrm{s}$ for a co-located 10-gigabit connection to its matcher. It also runs a microwave network from ZH4 to FR2, at 10 megabits a second, with equal access for all trading participants: an exchange selling the corridor to a neighbouring market’s building, as Euronext does from London. Frankfurt and Zurich are $306.7\,\mathrm{k}\mathrm{m}$ apart, $1.02\,\mathrm{m}\mathrm{s}$ in vacuum and $1.50\,\mathrm{m}\mathrm{s}$ in straight fibre.

11.3 The northern-Italy migration

Euronext’s frequently asked questions about the move, as it stood in November 2021, describe the old arrangement and the new one. Euronext relied on ICE’s Basildon data centre under a long-term service agreement; colocation was, in its press release’s word, “outsourced”. It chose a third-party building in Bergamo, in the country of Borsa Italiana, which it had just bought, to use the experience of the team that ran Borsa Italiana’s colocation. It kept its disaster-recovery site in the Paris region. It opened two points of presence in London, at InterXion and at Slough LD4, from which a client could reach Bergamo, the disaster-recovery site and Borsa Italiana’s Milan data centre, but not Basildon. And it opened the roof: wireless providers and trading firms could install antennas on a dedicated area of the roof, with Euronext guaranteeing equalisation from the back of the antenna to the meet-me room, the rule of chapter 9 applied to the air.

Figure 11.3 is the migration in numbers. From Slough the floor in vacuum went from $258\,\text{µ}\mathrm{s}$ to $3310\,\text{µ}\mathrm{s}$, $3.05\,\mathrm{m}\mathrm{s}$ more; in fibre from $377\,\text{µ}\mathrm{s}$ to $4839\,\text{µ}\mathrm{s}$, $4.46\,\mathrm{m}\mathrm{s}$ more. From Docklands, much the same. From Frankfurt the floor fell by $352\,\text{µ}\mathrm{s}$ (2.01 to $1.66\,\mathrm{m}\mathrm{s}$), and from Zurich by $1.81\,\mathrm{m}\mathrm{s}$. The London firm that had traded Euronext’s markets against London’s, a few hundred microseconds apart, now trades them against each other from opposite ends of a three-millisecond corridor, and chooses where to put the servers that react to each.

def migration(from_sites=("ld4", "docklands", "fr2", "zh4")):
    reg = registry()
    old, new = reg.primary("XPAR", on=BEFORE), reg.primary("XPAR", on=AFTER)
    rows = []
    for f in from_sites:
        d0, d1 = distance(f, old, reg), distance(f, new, reg)
        rows.append({"from": f, "old": old, "new": new, "km_before": d0 / 1e3, "km_after": d1 / 1e3,
                     "vac_before": gm.floor_us(d0), "vac_after": gm.floor_us(d1),
                     "fib_before": gm.floor_us(d0, "fibre"), "fib_after": gm.floor_us(d1, "fibre")})
    return rows

Before the migration, London, Frankfurt and Euronext formed a sliver: 678, 603 and 77 kilometres, with Euronext’s engine almost on top of London. After it, the triangle is wide: 678, 498 and 992 kilometres. A firm that relays between Frankfurt and Bergamo through London pays $678 + 992 = 1\,670$ kilometres of path for a direct distance of 498; a firm that wants to be between Euronext and Eurex should sit on the Frankfurt–Bergamo line, which is where Zurich almost lies (307 plus 205 kilometres, 511 against 498).

11.4 The Nordics and the rest

Nasdaq’s Nordic primary data centre stands in Upplands Väsby, 25 kilometres north of Stockholm, and its secondary moved in 2025 to Equinix SK2, south of the city. It is $1462.5\,\mathrm{k}\mathrm{m}$ from Slough ($4.88\,\mathrm{m}\mathrm{s}$ in vacuum) and $1195\,\mathrm{k}\mathrm{m}$ from Frankfurt: the farthest of the chapter’s sites from London, and a site to which chapters 22 and 23 return.

The rest of Europe follows the same pattern and the same method: each venue’s documents name a building, often in a carrier-neutral data centre in its own city; the registry holds one row per venue, role and period, with its source. Euronext’s 2024 release speaks of a single liquidity pool on one trading platform covering seven European markets, with its core data centre in Bergamo; the registry records there the six markets that the migration’s sources name, and leaves Borsa Italiana’s until a source dates its move.

    def rows(self, mic=None, role=None, on=None):
        keep = [r for r in self._rows if mic is None or r.mic == mic]
        return [r for r in keep if (role is None or r.role == role) and r.holds_on(on)]

    def primary(self, mic, on=None):
        found = {r.site for r in self.rows(mic, "primary", on)}
        if len(found) != 1:
            raise LookupError(f"{mic} on {on or 'today'}: {len(found)} primary sites")
        return found.pop()

    def venues_at(self, site_id, on=None):
        return sorted({r.mic for r in self.rows(role="primary", on=on) if r.site == site_id})

    def nearest(self, from_site, role="primary", on=None, medium="vacuum"):
        a = self.sites[from_site]
        out = []
        for r in self.rows(role=role, on=on):
            b = self.sites[r.site]
            d = gm.geodesic_m(a.lat, a.lon, b.lat, b.lon)
            out.append((r.mic, r.site, d / 1e3, gm.floor_us(d, medium)))
        return sorted(out, key=lambda x: (x[2], x[0]))

11.5 The London–Frankfurt corridor

The corridor’s floor is set by the geodesic: $677.7\,\mathrm{k}\mathrm{m}$ from Slough LD4 to Frankfurt FR2, $2.261\,\mathrm{m}\mathrm{s}$ in air and $3.305\,\mathrm{m}\mathrm{s}$ in straight fibre. In March 2015 McKay Brothers announced a route between the two buildings at $4.64\,\mathrm{m}\mathrm{s}$ round trip, $2.32\,\mathrm{m}\mathrm{s}$ one way: a route factor of 1.026 in air, $59\,\text{µ}\mathrm{s}$ above the floor, over a path that must cross the English Channel. It was already a millisecond faster than straight glass could ever be.

The London–Bergamo corridor is Euronext’s own. Its EWIN service, launched in July 2024 with McKay Brothers, carries orders by microwave from LD4 to IT3 in “less than 4 milliseconds”, with full fibre back-up. The floor in air is $3.311\,\mathrm{m}\mathrm{s}$, so the published bound allows a route factor of up to 1.21; straight fibre could not do better than $4.839\,\mathrm{m}\mathrm{s}$. A published bound is not a measurement: it tells a firm what the venue will stand behind, and the firm measures the rest (chapter 15).

    def holds_on(self, on):
        if on is None:
            return not self.valid_to
        started = not self.valid_from or self.valid_from <= on
        return started and (not self.valid_to or on <= self.valid_to)

11.6 Tutorial: the European registry

11.7 Build: the venue-to-site registry

11.8 Exercises

11.9 Problem: Basildon to Bergamo

11.10 Interview questions