vllm.distributed.eplb.migration_scheduler

Schedule EPLB expert migrations into conflict-free batches.

Classes:

• MigrationFlow –

  Expert transfers between one directed rank pair.

Functions:

• schedule_migration_batches –

  Schedule expert migrations without per-rank communication contention.

• schedule_migration_batches_for_layers –

  Precompute the independent migration schedule for every MoE layer.

MigrationFlow dataclass

Expert transfers between one directed rank pair.

Source code in vllm/distributed/eplb/migration_scheduler.py

@dataclass(frozen=True, slots=True)
class MigrationFlow:
    """Expert transfers between one directed rank pair."""

    src_rank: int
    dst_rank: int
    # Experts being transferred.
    expert_ids: tuple[int, ...]

schedule_migration_batches(num_local_experts, old_indices, new_indices)

Schedule expert migrations without per-rank communication contention.

Experts assigned to the same source and destination ranks form one flow. Each flow is placed in the first batch that does not already use either endpoint, so a rank communicates with at most one peer in each batch while independent rank pairs can transfer concurrently.

Source code in vllm/distributed/eplb/migration_scheduler.py

def schedule_migration_batches(
    num_local_experts: int,
    old_indices: np.ndarray,
    new_indices: np.ndarray,
) -> list[list[MigrationFlow]]:
    """Schedule expert migrations without per-rank communication contention.

    Experts assigned to the same source and destination ranks form one flow.
    Each flow is placed in the first batch that does not already use either
    endpoint, so a rank communicates with at most one peer in each batch while
    independent rank pairs can transfer concurrently.
    """
    assert old_indices.shape == new_indices.shape
    recv_ranks_by_expert: dict[int, list[int]] = {}
    old_experts_by_rank: list[set[int]] = []
    old_by_rank = old_indices.reshape(-1, num_local_experts)
    new_by_rank = new_indices.reshape(-1, num_local_experts)

    # First find the ranks that gain each expert. Existing local copies need no
    # transfer, even if their physical row changes.
    for rank, (old_local, new_local) in enumerate(zip(old_by_rank, new_by_rank)):
        old_experts = set(old_local.tolist())
        old_experts.discard(-1)
        old_experts_by_rank.append(old_experts)

        new_experts = set(new_local.tolist())
        new_experts.discard(-1)
        for expert_id in new_experts - old_experts:
            recv_ranks_by_expert.setdefault(expert_id, []).append(rank)

    old_ranks_by_expert: dict[int, list[int]] = {}
    needed_experts = recv_ranks_by_expert.keys()

    # Only index source copies for experts that need a remote transfer.
    for rank, old_experts in enumerate(old_experts_by_rank):
        for expert_id in old_experts & needed_experts:
            old_ranks_by_expert.setdefault(expert_id, []).append(rank)

    expert_ids_by_pair: dict[tuple[int, int], list[int]] = {}
    for expert_id in sorted(recv_ranks_by_expert):
        send_ranks = old_ranks_by_expert.get(expert_id, [])
        if not send_ranks:
            continue
        recv_ranks = recv_ranks_by_expert[expert_id]

        num_per_sender, remainder = divmod(len(recv_ranks), len(send_ranks))
        remainder_start = len(send_ranks) * num_per_sender

        # Preserve the unbatched path's balanced source assignment. Batching
        # changes transfer order, not which existing replica supplies a copy.
        for sender_idx, src_rank in enumerate(send_ranks):
            start = sender_idx * num_per_sender
            for dst_rank in recv_ranks[start : start + num_per_sender]:
                expert_ids_by_pair.setdefault((src_rank, dst_rank), []).append(
                    expert_id
                )
            if sender_idx < remainder:
                dst_rank = recv_ranks[remainder_start + sender_idx]
                expert_ids_by_pair.setdefault((src_rank, dst_rank), []).append(
                    expert_id
                )

    # Transfers with the same directed rank pair have identical contention
    # constraints, so coalesce them before grouping. The loop below is a
    # deterministic first-fit edge coloring of the rank communication graph.
    batches: list[list[MigrationFlow]] = []
    endpoints_used: list[int] = []
    for (src_rank, dst_rank), expert_ids in expert_ids_by_pair.items():
        flow = MigrationFlow(src_rank, dst_rank, tuple(expert_ids))

        # Direction is irrelevant for NIC contention. This endpoint mask keeps
        # each rank connected to at most one peer in a batch. The conservative
        # limit leaves bandwidth headroom for inference during async migration.
        endpoints = (1 << src_rank) | (1 << dst_rank)
        for batch_idx, (batch, used) in enumerate(zip(batches, endpoints_used)):
            if not endpoints & used:
                batch.append(flow)
                endpoints_used[batch_idx] |= endpoints
                break
        else:
            batches.append([flow])
            endpoints_used.append(endpoints)

    return batches

schedule_migration_batches_for_layers(num_local_experts, old_indices, new_indices)

Precompute the independent migration schedule for every MoE layer.

This reduces scheduling entry points from once per layer to once per rebalance cycle. Only CPU scheduling is grouped; transfers remain ordered by layer and preserve each layer's batch boundaries.

Source code in vllm/distributed/eplb/migration_scheduler.py

def schedule_migration_batches_for_layers(
    num_local_experts: int,
    old_indices: np.ndarray,
    new_indices: np.ndarray,
) -> list[list[list[MigrationFlow]]]:
    """Precompute the independent migration schedule for every MoE layer.

    This reduces scheduling entry points from once per layer to once per
    rebalance cycle. Only CPU scheduling is grouped; transfers remain ordered
    by layer and preserve each layer's batch boundaries.
    """
    assert old_indices.shape == new_indices.shape
    assert old_indices.ndim == 2
    return [
        schedule_migration_batches(num_local_experts, old_layer, new_layer)
        for old_layer, new_layer in zip(old_indices, new_indices)
    ]