Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Add argument to enable RMM alloaction tracking in benchmarks #1145

Merged
merged 1 commit into from
Mar 29, 2023
Merged

Add argument to enable RMM alloaction tracking in benchmarks #1145

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
1 change: 1 addition & 0 deletions dask_cuda/benchmarks/common.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -126,6 +126,7 @@ def run(client: Client, args: Namespace, config: Config):
args.rmm_release_threshold,
args.rmm_log_directory,
args.enable_rmm_statistics,
args.enable_rmm_track_allocations,
)
address_to_index, results, message_data = gather_bench_results(client, args, config)
p2p_bw = peer_to_peer_bandwidths(message_data, address_to_index)
Expand Down
20 changes: 20 additions & 0 deletions dask_cuda/benchmarks/utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -131,6 +131,17 @@ def parse_benchmark_args(description="Generic dask-cuda Benchmark", args_list=[]
"This enables spilling implementations such as JIT-Unspill to provides more "
"information on out-of-memory errors",
)
cluster_args.add_argument(
"--enable-rmm-track-allocations",
action="store_true",
help="When enabled, wraps the memory resource used by each worker with a "
"``rmm.mr.TrackingResourceAdaptor``, which tracks the amount of memory "
"allocated."
"NOTE: This option enables additional diagnostics to be collected and "
"reported by the Dask dashboard. However, there is significant overhead "
"associated with this and it should only be used for debugging and memory "
"profiling.",
)
cluster_args.add_argument(
"--enable-tcp-over-ucx",
default=None,
Expand Down Expand Up @@ -339,6 +350,7 @@ def get_cluster_options(args):
"CUDA_VISIBLE_DEVICES": args.devs,
"interface": args.interface,
"device_memory_limit": args.device_memory_limit,
"dashboard_address": 18787,
**ucx_options,
}
if args.no_silence_logs:
Expand Down Expand Up @@ -370,6 +382,7 @@ def setup_memory_pool(
release_threshold=None,
log_directory=None,
statistics=False,
rmm_track_allocations=False,
):
import cupy

Expand Down Expand Up @@ -399,6 +412,10 @@ def setup_memory_pool(
rmm.mr.set_current_device_resource(
rmm.mr.StatisticsResourceAdaptor(rmm.mr.get_current_device_resource())
)
if rmm_track_allocations:
rmm.mr.set_current_device_resource(
rmm.mr.TrackingResourceAdaptor(rmm.mr.get_current_device_resource())
)


def setup_memory_pools(
Expand All @@ -411,6 +428,7 @@ def setup_memory_pools(
release_threshold,
log_directory,
statistics,
rmm_track_allocations,
):
if not is_gpu:
return
Expand All @@ -423,6 +441,7 @@ def setup_memory_pools(
release_threshold=release_threshold,
log_directory=log_directory,
statistics=statistics,
rmm_track_allocations=rmm_track_allocations,
)
# Create an RMM pool on the scheduler due to occasional deserialization
# of CUDA objects. May cause issues with InfiniBand otherwise.
Expand All @@ -435,6 +454,7 @@ def setup_memory_pools(
release_threshold=release_threshold,
log_directory=log_directory,
statistics=statistics,
rmm_track_allocations=rmm_track_allocations,
)


Expand Down