feat: optimize _update_per_bucket_p2p logic

README.md

@@ -14,8 +14,8 @@ updating our [Kimi-K2](https://github.com/MoonshotAI/Kimi-K2) model (1 Trillion
 
 The core weight update logic is in `ParameterServer` class, a service colocated with inference engines. It provides two implementations of weight update: Broadcast and P2P.
 
-- **Broadcast**: Used when a large number of inference instances need to update weights in synchronous. This is the fastest implementation and should be used as the default update method. See `_update_per_bucket`.
-- **P2P**: Used when new inference instances are dynamically added (due to restarts or dynamic availability) while the existing instances are already serving requests. Under this scenario, to avoid affecting the workloads on existing instances, we use the [`mooncake-transfer-engine`](https://github.com/kvcache-ai/Mooncake?tab=readme-ov-file#use-python-package) to P2P send weights from CPUs in existing instances to GPUs in new instances. See `_update_per_bucket_p2p`.
+- **Broadcast**: Used when a large number of inference instances need to update weights in synchronous. This is the fastest implementation and should be used as the default update method. See `_update_per_bucket` with `ranks == None or []`.
+- **P2P**: Used when new inference instances are dynamically added (due to restarts or dynamic availability) while the existing instances are already serving requests. Under this scenario, to avoid affecting the workloads on existing instances, we use the [`mooncake-transfer-engine`](https://github.com/kvcache-ai/Mooncake?tab=readme-ov-file#use-python-package) to P2P send weights from CPUs in existing instances to GPUs in new instances. See `_update_per_bucket` with `ranks` specified.
 
 ### Optimized Weight Broadcast
 In the *Broadcast* implementation, the checkpoint-engine holds references to sharded weights in CPU memory, and need to efficiently broadcast them to a cluster of inference instances, often under a different sharding pattern.
@@ -36,16 +36,22 @@ It then executes the transfer, where it controls the inference engine through a
 
 Pipelining naturally requires more GPU memory. When memory is not enough, checkpoint-engine will fallback to serial execution.
 
+### Optimized P2P Bucket Assignment
+In the *P2P* implementation, checkpoint-engine needs to send weights from existing instances to new instances.
+To minimize the overall transfer time, checkpoint-engine optimizes the bucket assignment for each sender-receiver pair.
+The optimization goal is to make full use of the available network bandwidth for each sender and receiver.
+See [issue #25](https://github.com/MoonshotAI/checkpoint-engine/issues/25)
+
 ## Benchmark
 
 | Model                                | Device Info  | GatherMetas | Update (Broadcast) | Update (P2P)            |
 | :----------------------------------- | :----------- | :---------- |:-------------------| :---------------------- |
-| GLM-4.5-Air (BF16)                   | 8xH800 TP8  | 0.17s       | 3.94s (1.42GiB)    | 8.83s (4.77GiB)         |
-| Qwen3-235B-A22B-Instruct-2507 (BF16) | 8xH800 TP8  | 0.46s       | 6.75s (2.69GiB)    | 16.47s (4.05GiB)        |
-| DeepSeek-V3.1 (FP8)                  | 16xH20 TP16  | 1.44s       | 12.22s (2.38GiB)   | 25.77s (3.61GiB)        |
-| Kimi-K2-Instruct (FP8)               | 16xH20 TP16  | 1.81s       | 15.45s (2.93GiB)   | 36.24s (4.46GiB)        |
-| DeepSeek-V3.1 (FP8)                  | 256xH20 TP16 | 1.40s       | 13.88s (2.54GiB)   | 33.30s (3.86 GiB) |
-| Kimi-K2-Instruct (FP8)               | 256xH20 TP16 | 1.88s       | 21.50s (2.99GiB)   | 34.49s (4.57 GiB) |
+| GLM-4.5-Air (BF16)                   | 8xH800 TP8  | 0.12s       | 3.47s (1.42GiB)    | 4.12s (4.77GiB)         |
+| Qwen3-235B-A22B-Instruct-2507 (BF16) | 8xH800 TP8  | 0.33s       | 6.22s (2.69GiB)    | 7.10s (4.05GiB)        |
+| DeepSeek-V3.1 (FP8)                  | 16xH20 TP16  | 1.17s       | 10.46s (2.38GiB)   | 14.63s (3.61GiB)        |
+| Kimi-K2-Instruct (FP8)               | 16xH20 TP16  | 1.33s       | 14.51s (2.93GiB)   | 20.24s (4.46GiB)        |
+| DeepSeek-V3.1 (FP8)                  | 256xH20 TP16 | 0.94s       | 10.20s (2.54GiB)   | 13.82s (3.86 GiB) |
+| Kimi-K2-Instruct (FP8)               | 256xH20 TP16 | 1.24s       | 15.34s (2.99GiB)   | 19.69s (4.57 GiB) |
 
 All results above are tested by [`examples/update.py`](./examples/update.py) and use [vLLM v0.10.2rc1](https://github.com/vllm-project/vllm/tree/v0.10.2rc1) as inference engine. Some notes:
 
@@ -68,7 +74,7 @@ Use the flexible P2P implementation, notice this will install `mooncake-transfer
 pip install 'checkpoint-engine[p2p]'
 ```
 
-If set `NCCL_IB_HCA` env, checkpoint-engine will use it to auto select net devices for different ranks. If not set, it will read all RDMA devices and try to divide them into each rank.
+If set `NCCL_IB_HCA` env, checkpoint-engine will use it to auto select net devices for different ranks. Available patterns can be found from [NCCL documentation](https://docs.nvidia.com/deeplearning/nccl/user-guide/docs/env.html#id8). If not set, it will read all RDMA devices and try to divide them into each rank.
 
 ## Getting Started
 
@@ -145,7 +151,6 @@ torchrun --nproc-per-node 8 tests/test_update.py
 
 - This project is currently only tested with vLLM. But it is easy to integrate with other frameworks like SGLang.
 - The perfect three-stage pipeline mentioned in our paper is currently not implemented. This could be useful for architectures where H2D and broadcast do not conflict in PCIE.
-- The P2P update method is currently not the optimal implementation since it will receive data only in rank 0 and broadcast to others synchronizely. This is a potential optimization in the future.
 
 ## Acknowledgments