SGX & HKEX connectivity
Colocation, market data feeds, recovery services and the 2027 platform migrations in Singapore and Hong Kong.
9 questions ·quiz yourself → · knowledge base projected 2026-10-03
exchanges/sg-hk-exchanges.md · read the full note
Design: connect a new firm to SGX Iris-ST market data in the new PDC.
Cover cross-connects, primary and secondary feeds on separate paths, PIM-SM with SGX's RPs, recovery over Rewinder (multicast) and Glimpse (TCP), capture and timestamping, and how you would test in the Jan–Apr 2027 conformance window.
Why does HKEX say not to prefer line A?
Either line can arrive first, and the packets differ, so a feed handler must merge both by sequence number; "use B only on a gap on A" adds latency and loses data. See A/B arbitration in ../low-latency/07-ch07-transport.md.
Recovery design: why OMD-C's RTS limits (50,000 messages, 1,000 requests a day) push large-gap recovery to the refresh service.
The RTS is deliberately small: about 50,000 recent messages kept per stream, one connection per client, at most 1,000 requests a day and 10,000 messages per request, across four active servers. A large intraday gap blows through those caps, so clients take a refresh (multicast snapshot) instead and follow the live feed from the snapshot's sequence number.
How would you plan connectivity for the SGX Iris-ST data-centre migration?
Production IPs (and some ports) change for every service in the new primary data centre while SDC addresses stay; run the old and new data centres side by side over a high-capacity link during the migration; test access stays on the old site's QA handoff; non-colocated members move from MPLS towards IPsec VPN, and telcos migrate under SGX's program.
What replaces SGX REACH-ST, when is conformance testing, and what happens to production IP addresses?
Iris-ST; Jan–Apr 2027; all PDC production IPs change with the data centre move (SDC addresses stay).
Name SGX's two ITCH recovery mechanisms and when to use each.
Rewinder (multicast re-request of small gaps) and Glimpse (TCP snapshot after long disconnections).
What does HKEX's OMD-C guide say about listening to line B only when line A has a gap?
It is incorrect: listen to both lines with equal priority and arbitrate on sequence numbers.
Why should OMD-C clients not use the RTS to recover a large intraday gap?
The RTS keeps only about 50,000 recent messages per stream and limits requests; use the refresh service instead.
What does SGX FAQ 12.9 suggest about how SGX's multicast is built?
Market data uses rendezvous points, so PIM-SM with an RP (not SSM).