diff --git a/tests/e2e_test/test_e2e.py b/tests/e2e_test/test_e2e.py
index e8595223..11b8617f 100644
--- a/tests/e2e_test/test_e2e.py
+++ b/tests/e2e_test/test_e2e.py
@@ -156,7 +156,6 @@ async def test_e2e(clean_ray, model, migration_backend, launch_mode):
         "ignore_eos": False,
     }
 
-    launch_ray()
     # generate llumnix outputs
     base_port = 37037
     launch_llumnix_service(model, max_model_len, base_port, migration_backend, launch_mode)
diff --git a/tests/e2e_test/test_migration.py b/tests/e2e_test/test_migration.py
index 213d2637..b1f446f1 100644
--- a/tests/e2e_test/test_migration.py
+++ b/tests/e2e_test/test_migration.py
@@ -67,7 +67,7 @@ def parse_manager_log_file(log_file):
 @pytest.mark.asyncio
 @pytest.mark.skipif(torch.cuda.device_count() < 2, reason="at least 2 gpus required for migration bench")
 @pytest.mark.parametrize("model", ['/mnt/model/Qwen-7B'])
-@pytest.mark.parametrize("migration_backend", ['gloo'])
+@pytest.mark.parametrize("migration_backend", ['rpc', 'gloo'])
 async def test_migration_benchmark(clean_ray, model, migration_backend):
     base_port = 37037
     instance_output_logs = []