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

page-lock in spawn_blocking #893

Merged
merged 1 commit into from
Jan 10, 2025
Merged

page-lock in spawn_blocking #893

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
29 changes: 28 additions & 1 deletion iris-mpc-gpu/src/helpers/mod.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,10 +1,12 @@
use crate::threshold_ring::protocol::ChunkShare;
use cudarc::driver::{
result::{self, memcpy_dtoh_async, memcpy_htod_async, stream},
sys::{lib, CUdeviceptr, CUstream, CUstream_st},
sys::{lib, CUdeviceptr, CUstream, CUstream_st, CU_MEMHOSTALLOC_PORTABLE},
CudaDevice, CudaSlice, CudaStream, DevicePtr, DevicePtrMut, DeviceRepr, DriverError,
LaunchConfig,
};
use device_manager::DeviceManager;
use query_processor::CudaVec2DSlicerRawPointer;
use std::sync::Arc;

pub mod comm;
Expand Down Expand Up @@ -167,3 +169,28 @@ pub fn htod_on_stream_sync<T: DeviceRepr>(
};
Ok(buf)
}

pub fn register_host_memory(
device_manager: Arc<DeviceManager>,
db: &CudaVec2DSlicerRawPointer,
max_db_length: usize,
code_length: usize,
) {
let max_size = max_db_length / device_manager.device_count();
for (device_index, device) in device_manager.devices().iter().enumerate() {
device.bind_to_thread().unwrap();
unsafe {
let _ = cudarc::driver::sys::lib().cuMemHostRegister_v2(
db.limb_0[device_index] as *mut _,
max_size * code_length,
CU_MEMHOSTALLOC_PORTABLE,
);

let _ = cudarc::driver::sys::lib().cuMemHostRegister_v2(
db.limb_1[device_index] as *mut _,
max_size * code_length,
CU_MEMHOSTALLOC_PORTABLE,
);
}
}
}
1 change: 1 addition & 0 deletions iris-mpc-gpu/src/helpers/query_processor.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -68,6 +68,7 @@ impl<T> Drop for StreamAwareCudaSlice<T> {

/// Holds the raw memory pointers for the 2D slices.
/// Memory is not freed when the struct is dropped, but must be freed manually.
#[derive(Clone)]
pub struct CudaVec2DSlicerRawPointer {
pub limb_0: Vec<u64>,
pub limb_1: Vec<u64>,
Expand Down
72 changes: 36 additions & 36 deletions iris-mpc-gpu/src/server/actor.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -77,44 +77,44 @@ const KDF_SALT: &str = "111a1a93518f670e9bb0c2c68888e2beb9406d4c4ed571dc77b801e6
const SUPERMATCH_THRESHOLD: usize = 4_000;

pub struct ServerActor {
job_queue: mpsc::Receiver<ServerJob>,
device_manager: Arc<DeviceManager>,
party_id: usize,
job_queue: mpsc::Receiver<ServerJob>,
pub device_manager: Arc<DeviceManager>,
party_id: usize,
// engines
codes_engine: ShareDB,
masks_engine: ShareDB,
batch_codes_engine: ShareDB,
batch_masks_engine: ShareDB,
phase2: Circuits,
phase2_batch: Circuits,
distance_comparator: DistanceComparator,
comms: Vec<Arc<NcclComm>>,
codes_engine: ShareDB,
masks_engine: ShareDB,
batch_codes_engine: ShareDB,
batch_masks_engine: ShareDB,
phase2: Circuits,
phase2_batch: Circuits,
distance_comparator: DistanceComparator,
comms: Vec<Arc<NcclComm>>,
// DB slices
left_code_db_slices: SlicedProcessedDatabase,
left_mask_db_slices: SlicedProcessedDatabase,
right_code_db_slices: SlicedProcessedDatabase,
right_mask_db_slices: SlicedProcessedDatabase,
streams: Vec<Vec<CudaStream>>,
cublas_handles: Vec<Vec<CudaBlas>>,
results: Vec<CudaSlice<u32>>,
batch_results: Vec<CudaSlice<u32>>,
final_results: Vec<CudaSlice<u32>>,
db_match_list_left: Vec<CudaSlice<u64>>,
db_match_list_right: Vec<CudaSlice<u64>>,
batch_match_list_left: Vec<CudaSlice<u64>>,
batch_match_list_right: Vec<CudaSlice<u64>>,
current_db_sizes: Vec<usize>,
query_db_size: Vec<usize>,
max_batch_size: usize,
max_db_size: usize,
return_partial_results: bool,
disable_persistence: bool,
enable_debug_timing: bool,
code_chunk_buffers: Vec<DBChunkBuffers>,
mask_chunk_buffers: Vec<DBChunkBuffers>,
dot_events: Vec<Vec<CUevent>>,
exchange_events: Vec<Vec<CUevent>>,
phase2_events: Vec<Vec<CUevent>>,
pub left_code_db_slices: SlicedProcessedDatabase,
pub left_mask_db_slices: SlicedProcessedDatabase,
pub right_code_db_slices: SlicedProcessedDatabase,
pub right_mask_db_slices: SlicedProcessedDatabase,
streams: Vec<Vec<CudaStream>>,
cublas_handles: Vec<Vec<CudaBlas>>,
results: Vec<CudaSlice<u32>>,
batch_results: Vec<CudaSlice<u32>>,
final_results: Vec<CudaSlice<u32>>,
db_match_list_left: Vec<CudaSlice<u64>>,
db_match_list_right: Vec<CudaSlice<u64>>,
batch_match_list_left: Vec<CudaSlice<u64>>,
batch_match_list_right: Vec<CudaSlice<u64>>,
current_db_sizes: Vec<usize>,
query_db_size: Vec<usize>,
max_batch_size: usize,
max_db_size: usize,
return_partial_results: bool,
disable_persistence: bool,
enable_debug_timing: bool,
code_chunk_buffers: Vec<DBChunkBuffers>,
mask_chunk_buffers: Vec<DBChunkBuffers>,
dot_events: Vec<Vec<CUevent>>,
exchange_events: Vec<Vec<CUevent>>,
phase2_events: Vec<Vec<CUevent>>,
}

const NON_MATCH_ID: u32 = u32::MAX;
Expand Down
34 changes: 31 additions & 3 deletions iris-mpc/src/bin/server.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -31,9 +31,10 @@ use iris_mpc_common::{
sync::SyncState,
task_monitor::TaskMonitor,
},
IRIS_CODE_LENGTH, MASK_CODE_LENGTH,
};
use iris_mpc_gpu::{
helpers::device_manager::DeviceManager,
helpers::{device_manager::DeviceManager, register_host_memory},
server::{
get_dummy_shares_for_deletion, sync_nccl, BatchMetadata, BatchQuery,
BatchQueryEntriesPreprocessed, ServerActor, ServerJobResult,
Expand Down Expand Up @@ -1039,6 +1040,33 @@ async fn server_main(config: Config) -> eyre::Result<()> {
}
};

tracing::info!("Page-lock host memory");
let left_codes = actor.left_code_db_slices.code_gr.clone();
let right_codes = actor.right_code_db_slices.code_gr.clone();
let left_masks = actor.left_mask_db_slices.code_gr.clone();
let right_masks = actor.right_mask_db_slices.code_gr.clone();
let device_manager_clone = actor.device_manager.clone();

let page_lock_handle = spawn_blocking(move || {
for db in [&left_codes, &right_codes] {
register_host_memory(
device_manager_clone.clone(),
db,
config.max_db_size,
IRIS_CODE_LENGTH,
);
}

for db in [&left_masks, &right_masks] {
register_host_memory(
device_manager_clone.clone(),
db,
config.max_db_size,
MASK_CODE_LENGTH,
);
}
});

let now = Instant::now();
let mut now_load_summary = Instant::now();
let mut time_waiting_for_stream = time::Duration::from_secs(0);
Expand Down Expand Up @@ -1133,8 +1161,8 @@ async fn server_main(config: Config) -> eyre::Result<()> {
tracing::info!("Preprocessing db");
actor.preprocess_db();

tracing::info!("Page-lock host memory");
actor.register_host_memory();
tracing::info!("Waiting for page-lock to finish");
page_lock_handle.await?;

tracing::info!(
"Loaded {} records from db into memory [DB sizes: {:?}]",
Expand Down
Loading