MoE Serving (Wide EP)

MoE serving is primarily a design problem in token routing, expert imbalance, and collective-communication tails—not parameter count.

Co-design batching, routing, all-to-all communication, and failure scope for wide expert parallel deployments.

§ 01

Problem definition

The operating conditions that justify MoE Serving (Wide EP)

Even with balanced averages, token concentration on a few experts makes every rank wait for the slowest.

Adding GPUs distributes memory while expanding communication and failure domains.

  • The total model is large relative to active parameters and single-node placement is inefficient
  • Expert-parallel width can be validated against real token distributions
  • Network topology and collectives cannot be controlled
  • At low traffic, complexity costs exceed memory benefits
PLATE 01

MoE Serving (Wide EP): system plate

CONTROL

System 1

Profile expert selection and token imbalance by prompt class.

CONTROL

System 2

Compare intra-node and inter-node EP/TP placement against physical topology.

EXECUTION

System 3

Measure all-to-all volume, overlappable compute, and straggler impact by stage.

EXECUTION

System 4

Validate request scope and recovery strategy under expert, rank, and link failures.

  1. N1 N2context
  2. N2 N3decision
  3. N3 N4evidence
A decision and validation view for MoE Serving (Wide EP); labels describe architecture, not a measured deployment result.
  1. The workflow begins with Profile expert selection and token imbalance by prompt class..
  2. It reaches an acceptance decision through Expert load balance.

§ 03

Design method

Fix the boundary and acceptance criteria before implementation.

MoE serving is primarily a design problem in token routing, expert imbalance, and collective-communication tails—not parameter count.

  1. 01

    Stage 1

    Profile expert selection and token imbalance by prompt class.

    Review artifact 1
  2. 02

    Stage 2

    Compare intra-node and inter-node EP/TP placement against physical topology.

    Review artifact 2
  3. 03

    Stage 3

    Measure all-to-all volume, overlappable compute, and straggler impact by stage.

    Review artifact 3
  4. 04

    Stage 4

    Validate request scope and recovery strategy under expert, rank, and link failures.

    Review artifact 4

§ 04

Application scenarios

Hypothetical workloads make the applicability boundary concrete.

Hypothetical application scenario

The total model is large relative to active parameters and single-node placement is inefficient

Even with balanced averages, token concentration on a few experts makes every rank wait for the slowest.

APPROACH
Profile expert selection and token imbalance by prompt class.
BOUNDARY
Throughput figures are not transferred across different hardware, models, or traffic.
Hypothetical application scenario

Expert-parallel width can be validated against real token distributions

Adding GPUs distributes memory while expanding communication and failure domains.

APPROACH
Compare intra-node and inter-node EP/TP placement against physical topology.
BOUNDARY
Wide EP is not universally superior to TP or narrower EP.

§ 05

Design choices

Review gains and costs in the same table.

DecisionGainCostWatch
The total model is large relative to active parameters and single-node placement is inefficientProfile expert selection and token imbalance by prompt class.Throughput figures are not transferred across different hardware, models, or traffic.Expert load balance
Expert-parallel width can be validated against real token distributionsCompare intra-node and inter-node EP/TP placement against physical topology.Wide EP is not universally superior to TP or narrower EP.Collective share and tail time
PLATE 02

MoE Serving (Wide EP): system plate

ItemMethodEvidenceBoundary
Layer 1Profile expert selection and token imbalance by prompt class.Expert load balanceThroughput figures are not transferred across different hardware, models, or traffic.
Layer 2Compare intra-node and inter-node EP/TP placement against physical topology.Collective share and tail timeWide EP is not universally superior to TP or narrower EP.
Layer 3Measure all-to-all volume, overlappable compute, and straggler impact by stage.Token goodput and recovery scopeThroughput figures are not transferred across different hardware, models, or traffic.
A decision and validation view for MoE Serving (Wide EP); labels describe architecture, not a measured deployment result.
  1. The workflow begins with Profile expert selection and token imbalance by prompt class..
  2. It reaches an acceptance decision through Expert load balance.

§ 07

Validation plan

Agree on measurement conditions before publishing a result.

MeasureMethodPass conditionCaveat
Expert load balanceProfile expert selection and token imbalance by prompt class.Repeated runs satisfy the acceptance threshold agreed during discoveryThroughput figures are not transferred across different hardware, models, or traffic.
Collective share and tail timeCompare intra-node and inter-node EP/TP placement against physical topology.Repeated runs satisfy the acceptance threshold agreed during discovery
Token goodput and recovery scopeMeasure all-to-all volume, overlappable compute, and straggler impact by stage.Repeated runs satisfy the acceptance threshold agreed during discovery

§ 08

Constraints and failure conditions

Conditions for not applying the capability are part of the design.

Network topology and collectives cannot be controlled

Throughput figures are not transferred across different hardware, models, or traffic.

At low traffic, complexity costs exceed memory benefits

Wide EP is not universally superior to TP or narrower EP.

§ 10

Durable deliverables

Artifacts remain with the operating organization after the engagement.

MoE Serving (Wide EP) decision record
MoE serving is primarily a design problem in token routing, expert imbalance, and collective-communication tails—not parameter count.Client-owned · Patty-reviewed
Validation harness and acceptance criteria
Expert load balance · Collective share and tail time · Token goodput and recovery scopeJointly maintained
Operations and recovery runbook
Throughput figures are not transferred across different hardware, models, or traffic. · Wide EP is not universally superior to TP or narrower EP.Operating-team owned

§ 11

Terminology

Use shared terms with explicit operating meaning.

MoE Serving (Wide EP)
Co-design batching, routing, all-to-all communication, and failure scope for wide expert parallel deployments.
Acceptance criterion
Expert load balance
Operating boundary
Throughput figures are not transferred across different hardware, models, or traffic.

REFERENCES

References and primary material

  1. DeepEP

    Primary material for the method and terminology.

  2. vLLM Expert Parallel Deployment

    Primary material for the method and terminology.

Begin by determining whether MoE Serving (Wide EP) is the justified next step.

We define scope and validation against representative work, data and infrastructure boundaries, and explicit failure conditions.

Request a technical review