Skip to content
QuantLabTickDirect

Ultra-low-latency FX rate feed

Tick servers pass each FX rate from the liquidity provider (LP) to every subscriber within 0.21 ms (p99). They sit next to the LP's distribution points in Tokyo, Singapore, London, and New York.

Beta service with no SLA. Plans start at 100 USDC per 30 days.

Relay latency

0.21msp99

From a Tick server receiving a rate to sending it to every subscriber

Load test delivering 600,000 ticks/s (200 subscribers)

p50 0.12 ms, p99 0.21 ms

Round trip to the LP

  • Tokyo0.3 ms
  • Singapore1.0 ms
  • London0.4 ms
  • New York1.8 ms

Measured values. See the notes at the end of this page.

Why it's fast

The transport, the data format, where the servers run, and how they send were each chosen to get rates to you sooner.

UDP: no waiting on lost data

With TCP, one lost packet holds back everything behind it until the packet is resent and arrives. A rate is replaced by the next update, so a new rate now is worth more than a resent old one. TickDirect sends over UDP and does not resend lost datagrams; sequence numbers tell you when one is missing.

TCP
12342 3 4

3 and 4 wait until 2 is resent

UDP
1234

3 and 4 are available as they arrive

When packet 2 is losttime →

FIX SBE: values you can read as they arrive

SBE (Simple Binary Encoding) is a binary format for low-latency use, standardized by FIX Trading Community. Every value has a fixed position and length, so there is no text to parse and convert as with JSON. The bid is the 8 bytes at a fixed offset, read directly as an integer. You can also generate code for your language from the XML schema.

symbolId
from byte 0
2 bytes
bid
from byte 2
8 bytes
ask
from byte 10
8 bytes
upstreamTime
from byte 18
8 bytes
receiveTime
from byte 26
8 bytes
One tick = 34 bytes

Placed next to the LP

We measured the round trip from each candidate AWS Availability Zone to the LP's distribution point and put each Tick server in the closest one. In New York, for example, the server runs in the New York Local Zone (1.8 ms round trip) rather than the usual North Virginia Region (about 6.5 ms).

  • AWS North Virginia (us-east-1)

    ~6.5 ms
  • AWS New York Local Zone (TickDirect)

    1.8 ms
Round trip to the LP's distribution point in New York

3 million ticks per second, under 1 ms at p99

In our load test, each subscriber received 30 symbols × 100 ticks/s (3,000 ticks/s), and we added subscribers up to 1,000. Delivering 3 million ticks per second in total, the time from receiving a rate to sending it to the last subscriber was 0.85 ms at p99. Tick servers run on a custom-tuned OS that bypasses the operating system's network stack and sends directly to the NIC with zero copy.

  • 600,000 ticks/s (200 subscribers)

    0.21 ms
  • 1.2 million ticks/s (400 subscribers)

    0.38 ms
  • 3 million ticks/s (1,000 subscribers)

    0.85 ms
Ticks delivered per second and relay latency (p99, load test)

Easy to adopt

Everything you need to go from a subscription to rates in your own system.

  • Integer prices, UTC nanosecond timestamps

    Prices arrive as integers scaled by 10⁸ (1.08529 is 108529000), so you can compare and subtract them without floating-point rounding. Timestamps are UTC nanoseconds; the Tick server normalizes each LP's time zone before sending.

  • Measure delivery time yourself

    Each tick carries the time the Tick server received the rate. Compare it with the time your system received the tick to see how long delivery took.

  • Subscribe to just what you need

    Choose up to 64 symbols by name. You can change the set at any time without reconnecting.

  • Four locations on the 1 IP plan

    IP addresses are counted per Tick server. Even with the 1 IP plan, you can connect to Tokyo, Singapore, London, and New York at the same time, from one IP address each.

  • Works behind NAT

    Tick servers send to the address a client's requests came from, so you can receive behind NAT or a firewall without opening an inbound port.

  • Docs in English and Japanese, and a Python sample

    The protocol is documented in English and Japanese. A single-file Python sample takes you from authentication to receiving rates.

Locations

Rates are served from four locations. Our contracts do not allow us to disclose the name of the rate source (the LP), so each LP has a codename. Today there is one LP, “Alpha”, and we plan to add more.

  • Tokyo

    AWS Tokyo Region

    Round trip to the LP

    0.3ms

    18 symbols

  • Singapore

    AWS Singapore Region

    Round trip to the LP

    1.0ms

    18 symbols

  • London

    AWS London Region

    Round trip to the LP

    0.4ms

    20 symbols

  • New York

    AWS New York Local Zone

    Round trip to the LP

    1.8ms

    11 symbols

Locations, the LP, and symbols by location

Getting started

Three steps from subscribing to receiving rates.

  1. 1

    Choose a plan

    On ggguilds, pick the plan that matches how many IP addresses you will connect from.

  2. 2

    Issue an API key

    Once your first payment is confirmed, issue a key on the contract page. It works within about a minute.

  3. 3

    Connect

    Check that it works with the Python sample, then build it into your system with the protocol docs.

FAQ

Things to check before you subscribe.

Can data be lost?

Datagrams can be lost on the network because they are sent over UDP. Lost rates are not resent, but the next update brings the latest value, and sequence numbers tell you when something is missing.

Do I get the current rate when I subscribe?

No. Rates arrive from the next update after you subscribe.

Who is the LP (the rate source)?

Our contracts do not allow us to disclose LP names. Each LP has a codename; today there is one LP, “Alpha”. We plan to add more LPs.

Which programming languages can I use?

Any language that can send and receive UDP. There is a Python sample, and the SBE schema can generate code for other languages.

Is the connection encrypted?

Rates are sent unencrypted. Authentication uses a JWT signed with your API key.

Can I connect over IPv6?

No, only IPv4 is supported.

Is there an SLA?

No. This is a beta service with no SLA.

What happens if I cancel or reissue my API key?

After you cancel, you keep access until the end of the paid period. When the period ends or you reissue your key, delivery on the old key stops, even on open connections.

Read the protocol

How we measured

  • Relay latency is the time from when a Tick server receives a rate until it has sent the rate to the last subscriber. It was measured in a load test on the same server type as production (AWS c6in.large). A mock upstream stood in for the LP and generated 30 symbols × 100 ticks/s. Each of 200, 400, and 1,000 subscribers, subscribed to every symbol, received 3,000 ticks/s, for 600,000, 1.2 million, and 3 million ticks delivered per second in total. The real LP alpha sends fewer ticks than this. It does not include time on the network. With 200 subscribers, the p50 is 0.12 ms.
  • The round trip to the LP is the median TCP connection time from each Tick server's Availability Zone to the LP's distribution point (measured on November 17, 2026). The North Virginia figure was measured the same way from us-east-1 Availability Zones.
  • None of these values is guaranteed.