You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Copy file name to clipboardExpand all lines: examples/python/remote_storage_example/README.md
+18Lines changed: 18 additions & 0 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -108,3 +108,21 @@ The system automatically selects the best available storage backend:
108
108
1. Initiator sends memory descriptors to target
109
109
2. Target performs storage-to-memory or memory-to-storage operations
110
110
3. Data is transferred between initiator and target memory
111
+
112
+
Remote reads are implemented as a read from storage followed by a network write.
113
+
114
+
Remote writes are implemented as a read from network following by a storage write.
115
+
116
+
### Pipelining
117
+
118
+
To improve performance of the remote storage server, we can pipeline operations to network and storage. This pipelining allows multiple threads to handle each request. However, in order to maintain correctness, the order of network and storage must happen in order for each individual remote storage operation. To do this, we implemented a simple pipelining scheme.
For high-speed storage and network hardware, you may need to tweak performance with a couple of environment variables.
125
+
126
+
First, for optimal GDS performance, ensure you are using the GDS_MT backend with default concurrency. Additionally, you can use the cufile options described in the [GDS README](https://github.com/ai-dynamo/nixl/blob/main/src/plugins/cuda_gds/README.md).
127
+
128
+
On the network side, remote reads from VRAM to DRAM can be limited by UCX rail selection. This can be tweaked by setting UCX_MAX_RMA_RAILS=1. However, with larger batch or message sizes, this might limit bandwidth and a higher number of rails might be needed.
0 commit comments