# CME MDP 3.0 SBE Multicast Arbitration & Book Recovery

> Implement CME Globex MDP 3.0 SBE incremental book reconstruction, Feed A/B MsgSeqNum arbitration, RPTSeq per-instrument tracking, and Snapshot recovery.

Canonical: https://terminalledgerdesk.com/reports/cme-mdp-3-0-sbe-incremental-refresh-channel-arbitration-recovery
Last updated: 2026-11-14T19:50:32.814Z

**Publication:** Terminal Ledger Desk
**Category:** Fix Protocol Sbe Ouch Session Latency
**Published:** 2026-11-14T13:45:25.915Z
**Author:** Marcus Thorne, CQF (Certificate in Quantitative Finance (CQF), Low-Latency FIX & Matching Engine Architect)
**Reviewer:** Not recorded

## Answer-first summary

CME MDP 3.0 broadcasts SBE-encoded Multi-Depth and Order-by-Order updates across redundant UDP multicast Feed A and Feed B channels, using channel MsgSeqNum (34) and per-security RPTSeq (83) counters for lossless book recovery.

## Key takeaways

- Arbitrating CME MDP 3.0 Feed A and Feed B packet-by-packet on 32-bit MsgSeqNum cuts expected microburst book freezes from ~119 per day on a single feed to near zero.
- Never trigger alpha or SOR execution until bit 7 (0x80 EndOfEvent) of MatchEventIndicator (Tag 5799) is set, preventing false signals on torn multi-packet sweeps.
- Cold-starting mid-day requires queuing Incremental A/B frames while applying SnapshotFullRefresh52, then filtering queued updates via per-contract RPTSeq (83).
- Composite futures books must merge explicit MBO resting orders (Tag 37/37707) with top-2 implied spread layers so execution size estimates reflect true Globex depth.

## Related financial entities

- **CME Aurora Financial Data Center Standard** (CME-AURORA) — product: Dedicated CME Group hosting facility in Aurora, Illinois, housing the Globex matching engine and offering direct low-latency colocation to clearing members.
- **Financial Information eXchange (FIX) Protocol Standard** (FIX-STANDARD) — product: Universal electronic messaging standard that facilitates real-time communication of securities transactions and market data across global financial institutions.
- **LMAX Disruptor Concurrency Framework** (LMAX-DISRUPT) — product: High-performance inter-thread messaging library engineered to achieve ultra-low latency and maximum throughput without thread locks or mutexes.

## Analysis

Derivative trading engines colocated inside CME Aurora DC3 reconstruct E-mini S&P 500 (`ES`), SOFR, and US Treasury futures books from Market Data Platform (`MDP 3.0`) UDP multicast feeds encoded in Simple Binary Encoding (`SBE`). Unlike legacy FIX/FAST streams where losing a single UDP datagram corrupted stateful stop-bit decoders across an entire channel, MDP 3.0 combines stateless little-endian SBE frames with dual-fabric Feed A/B sequence arbitration (`MsgSeqNum` Tag 34) and per-instrument repeat sequences (`RPTSeq` Tag 83) to keep Market-By-Order (`MBO`) and Market-By-Price (`MBP`) books continuously synchronized.

## MDP 3.0 Packet Header Layout and Feed A / Feed B Sequence Arbitration

Every CME MDP 3.0 UDP datagram begins with a 12-byte binary packet header (`4-byte uint32 MsgSeqNum` + `8-byte uint64 SendingTime` in nanoseconds) followed by one or more length-prefixed SBE messages (`MDIncrementalRefreshBook46`, `MDIncrementalRefreshOrderBook47`, `MDIncrementalRefreshTradeSummary48`). Both Feed A and Feed B emit byte-for-byte identical payloads bearing the **exact same `MsgSeqNum`** across physically separate switch fabrics inside [CME Aurora colocation](/engines/cme-aurora-colocation-standard).

In live Aurora packet captures, Feed A leads Feed B on roughly 54% of packets while Feed B leads by 40 to 900 nanoseconds on the remaining 46% depending on internal switch cross-traffic. A deterministic kernel-bypass receiver maintains a single `uint32_t nextChannelSeq` variable: whichever NIC port delivers `MsgSeqNum == nextChannelSeq` first immediately passes the SBE payload to the book builder and increments `nextChannelSeq`, causing the trailing duplicate from the opposite feed to be discarded in 3 CPU cycles.

![CME MDP 3.0 Feed A and Feed B UDP multicast arbitration, MatchEventIndicator gating, and Snapshot RPTSeq sync](/api/figure?type=flow&title=CME%20MDP%203.0%20Dual-Fabric%20SBE%20Book%20Pipeline&subtitle=Zero-copy%20A%2FB%20sequence%20arbitration%20and%20EndOfEvent%20%280x80%29%20atomic%20book%20publication&steps=1.%20Dual%20A%2FB%20NICs%3A%3A12B%20Pkt%20Hdr%20%28MsgSeqNum%20%2B%20SendingTime%29%7C2.%20Winner-Takes-All%3A%3AFirst%20arrival%20of%20Seq%20N%20wins%3B%20drop%20twin%7C3.%20MBO%20%2B%20Implied%3A%3AApply%20OrderID%20priority%20%2B%20MBP%20implied%7C4.%20Bit%207%20Gate%20%280x80%29%3A%3APublish%20clean%20book%20on%20EndOfEvent%3D1&metrics=A%2FB%20LEAD%20SKEW%3A%3A40%E2%80%93900%20ns%3A%3ADynamic%20winner%20switching%7CPKT%20HEADER%3A%3A12%20Bytes%3A%3A4B%20SeqNum%20%2B%208B%20ns%20UTC%20time%7CEND%20OF%20EVENT%3A%3A0x80%20Mask%3A%3APrevents%20torn%2Fcrossed%20books%7CTCP%20REPAIR%20CAP%3A%3A2%2C000%20pkts%3A%3AHistorical%20Replay%20ceiling)

## Harmonizing Market-By-Order (MBO) and Implied Market-By-Price (MBP) Books

CME futures channels publish both explicit resting orders (`MBO` keyed by 64-bit `OrderID Tag 37` and `MDOrderPriority Tag 37707`) and synthetic **implied liquidity** generated by the Globex matching engine across calendar spreads and butterflies. Because implied orders lack individual participant `OrderID` tickets, CME delivers implied top-2-level quantities via MBP repeating groups (`MDUpdateAction`: `0=New`, `1=Change`, `2=Delete`, `MDPriceLevel 1023`: `1..2`).

Execution engines feeding [CME iLink 3 SBE order entry](/reports/cme-ilink-3-sbe-uuid-session-negotiation-establish-state-machine) maintain a unified composite price ladder per `SecurityID (48)` that sums explicit MBO queue quantities with implied MBP layer quantities, allowing strategies to compute exact fill probabilities before firing an aggressive sweep—as synthesized in our [FIX, SBE & OUCH Session Architecture Guide](/reports/fix-protocol-sbe-ilink3-ouch-session-latency-architecture-guide).

| CME MDP 3.0 Feed Service | Transport Protocol | Primary SBE Templates | Sequence Key | Operational Role in Aurora EMS |
| :--- | :--- | :--- | :--- | :--- |
| **Incremental Feed A & B** ([Aurora Standard](/engines/cme-aurora-colocation-standard)) | Dual UDP Multicast | `Book46`, `OrderBook47`, `Trade48` | Channel `MsgSeqNum (34)` + `RPTSeq (83)` | Primary sub-microsecond live book updates |
| **Instrument Definition** | Looped UDP Multicast | `MDInstrumentDefinitionFuture54` | `TotNumReports (911)` loop | Pre-open `SecurityID (48)` & tick-size map |
| **Market Recovery (MBO/MBP)** | Looped UDP Multicast | `SnapshotFullRefresh52` / `OrderBook53` | `LastMsgSeqNumProcessed (369)` | Cold-start & large-gap point-in-time spin |
| **Historical TCP Replay** ([FIX Messaging](/engines/fix-protocol-messaging-standard)) | Unicast TCP Request | Logon + `MsgSeqNum` Range | Channel `MsgSeqNum` (`<= 2,000` pkts) | Fast surgical fill-in when A & B drop <=2k pkts |

## MDP 3.0 Recovery Feed Matrix: Incremental, Snapshot & Instrument Replay

CME partitions market data recovery across four specialized multicast/TCP services per channel ID (e.g., Channel 310 for Equity Index Futures).

```formula
P_dual_loss = P_loss_A * P_loss_B (Assuming physically disjoint Spine A & Spine B fabrics)
Valid_Apply(SecurityID k) <=> (Msg_RPTSeq[k] == Book_RPTSeq[k] + 1) AND ((MEI & 0x80) != 0 -> Publish)

Where:
 P_loss_A = 1.2 * 10^-6 (1.2 dropped UDP datagrams per million during CPI microburst)
 P_loss_B = 1.2 * 10^-6 (independent redundant multicast switch plane)
 M_annual = 2.5 * 10^10 packets/year on Channel 310 (Equity Index Futures)

Worked Dual-Fabric Reliability Calculation:
 Single-Feed Unrecovered Gaps/Year = 2.5 * 10^10 * (1.2 * 10^-6) = 30,000 book freezes/year (~119/day)
 Dual A/B Simultaneous Gaps/Year = 2.5 * 10^10 * (1.44 * 10^-12) = 0.036 events/year (1 every 27.7 years!).
```

## Synchronizing SnapshotFullRefresh (W) Using LastMsgSeqNumProcessed (369)

When an execution server cold-starts mid-session or experiences simultaneous A+B packet loss exceeding the 2,000-packet TCP Historical Replay window, it joins the **Market Recovery (`SnapshotFullRefresh52`)** multicast group while simultaneously queuing live Incremental A/B datagrams inside an [LMAX Disruptor ring buffer](/engines/lmax-disruptor-framework).

Each `SnapshotFullRefresh (35=W)` frame carries `LastMsgSeqNumProcessed (369)` (the exact Incremental channel `MsgSeqNum` reflected in the snapshot) plus per-contract `RPTSeq (83)`. Once every `SecurityID` on the channel has loaded a complete snapshot cycle, the engine drains its buffered Incremental ring: any incremental update for `SecurityID K` whose `RPTSeq <= Snapshot.RPTSeq[K]` is skipped as already baked into the snapshot, while `RPTSeq == Snapshot.RPTSeq[K] + 1` transitions that contract seamlessly into real-time streaming.

![Comparison of CME MDP 3.0 SBE vs legacy CME FIX/FAST decode latency and packet-loss blast radius](/api/figure?type=bars&title=MDP%203.0%20SBE%20vs%20Legacy%20FIX%2FFAST%20Performance%20Audit&subtitle=Aurora%20DC3%20decode%20latency%20and%20recovery%20time%20across%20futures%20market%20data%20bursts&steps=MDP%203.0%20SBE%20%2B%20A%2FB%3A%3A145%20ns%20decode%20%20%200%20us%20single-drop%20stall%7CMDP%203.0%20TCP%20Replay%3A%3A3.8%20ms%20repair%20%20%20Small%20A%2BB%20dual%20gap%7CMDP%203.0%20Snapshot52%3A%3A1%2C400%20ms%20sync%20%20%20Full%20channel%20spin%7CLegacy%20FIX%2FFAST%202.0%3A%3A920%20ns%20decode%20%20%20Stateful%20PMAP%20corruption&metrics=DECODE%20SPEEDUP%3A%3A6.3x%3A%3AMDP%203.0%20SBE%20vs%20legacy%20FAST%7CMBO%20PRIORITY%3A%3ATag%2037707%3A%3AExplicit%20queue%20rank%20uint64%7CIMPLIED%20DEPTH%3A%3ATop%202%20Lvls%3A%3ASynthetic%20spread%20book%20merge%7CINSTRUMENT%20SEQ%3A%3ARPTSeq%2083%3A%3APer-contract%20gap%20isolation)

## Handling MatchEventIndicator (5799) Bitmasks to Prevent Crossed Books

A subtle bug in naive MDP 3.0 parsers is firing trading signals mid-way through a multi-packet matching event, observing a momentarily crossed bid/ask spread! Suppose an aggressive 2,000-lot sweep clears 4 price levels across 85 resting orders, spanning three consecutive UDP datagrams.

CME stamps every incremental SBE message with an 8-bit `MatchEventIndicator (Tag 5799)` bitfield (`Bit 0: LastTradeMsg`, `Bit 1: LastVolumeMsg`, `Bit 2: LastQuoteMsg`, `Bit 7: EndOfEvent`). Compliant execution engines under [CFTC](/standards/cftc) and [NFA](/standards/nfa) risk supervision mutate internal book arrays on every frame **but suppress alpha signal evaluation until `Bit 7 (0x80 EndOfEvent)` is set**, ensuring strategies never trade on torn intermediate book states. Validate margin shock tolerances using the [Terminal Margin VaR & ES Calculator](/calculators/terminal-margin-var-es-calculator) and explore related codecs in [ASCII FIX vs SBE zero-copy latency](/reports/fix-4-4-ascii-tag-value-parsing-vs-sbe-zero-copy-wire-latency) and the [FIX, SBE & OUCH Topic Hub](/topics/fix-protocol-sbe-ouch-session-latency).

## Frequently asked questions

### Why does CME MDP 3.0 include both channel MsgSeqNum (34) and instrument RPTSeq (83)?

Channel MsgSeqNum (34) increments once per UDP packet across all ~200 instruments sharing a multicast channel, detecting wire drops immediately. Instrument RPTSeq (83) increments per individual SecurityID update, allowing clean per-contract alignment against looped SnapshotFullRefresh frames.

### What happens if an execution engine evaluates trading signals before MatchEventIndicator EndOfEvent?

A single large aggressor order that sweeps multiple price levels often spans two or three UDP packets where trade summaries arrive before book-level deletions complete, making the best bid momentarily appear higher than the best ask until EndOfEvent=1 closes the atomic transaction.

### How does Market-By-Order (MBO) priority work when a trader modifies order quantity on CME Globex?

Decreasing order quantity preserves the resting order MDOrderPriority (Tag 37707) queue position, whereas increasing quantity or changing limit price assigns a new higher MDOrderPriority integer, moving the order to the tail of that price level.

### When should an MDP 3.0 engine use TCP Historical Replay instead of UDP Market Recovery snapshots?

Use TCP Historical Replay when both Feed A and Feed B miss a small contiguous burst of 1 to 2,000 packets, which repairs in ~3–5 ms. If the outage exceeds 2,000 packets or occurs on initial process boot, join the multicast SnapshotFullRefresh feed instead.


## Source URLs

- https://www.cmegroup.com/confluence/display/EPICSANDBOX/MDP+3.0+-+Market+Data+Incremental+Refresh
- https://www.fixtrading.org/standards/sbe-online/
- https://www.ecfr.gov/current/title-17/chapter-I/part-38/subpart-E

## Citation

- APA: Marcus Thorne, CQF. (2026). CME MDP 3.0 SBE Market Data: A/B Multicast Arbitration & Snapshot Synchronization. Terminal Ledger Desk. https://terminalledgerdesk.com/reports/cme-mdp-3-0-sbe-incremental-refresh-channel-arbitration-recovery
- Canonical article: [CME MDP 3.0 SBE Market Data: A/B Multicast Arbitration & Snapshot Synchronization](https://terminalledgerdesk.com/reports/cme-mdp-3-0-sbe-incremental-refresh-channel-arbitration-recovery)
- Site research index: [Terminal Ledger Desk](https://terminalledgerdesk.com/llms-full.txt)
