Designing MPI and PGAS Libraries for Exascale Systems: The MVAPICH2 Approach

Transcription

1 Designing MPI and PGAS Libraries for Exascale Systems: The MVAPICH2 Approach Talk at OpenFabrics Workshop (March 217) by Dhabaleswar K. (DK) Panda The Ohio State University

2 OFA (March 17) 2 High-End Computing (HEC): Towards Exascale-Era 15-3 PFlops in ? 1 EFlops in 221? Expected to have an ExaFlop system in 221!

3 Drivers of Modern HPC Cluster Architectures Multi-core Processors High Performance Interconnects - InfiniBand <1usec latency, 1Gbps Bandwidth> Multi-core/many-core technologies Remote Direct Memory Access (RDMA)-enabled networking (InfiniBand and RoCE) Solid State Drives (SSDs), Non-Volatile Random-Access Memory (NVRAM), NVMe-SSD Accelerators (NVIDIA GPGPUs and Intel Xeon Phi) Accelerators / Coprocessors high compute density, high performance/watt >1 TFlop DP on a chip Available on HPC Clouds, e.g., Amazon EC2, NSF Chameleon, Microsoft Azure, etc. SSD, NVMe-SSD, NVRAM Sunway TaihuLight K - Computer Tianhe 2 Titan OFA (March 17) 3

4 Three Major Computing Categories Scientific Computing Message Passing Interface (MPI), including MPI + OpenMP, is the Dominant Programming Model Many discussions towards Partitioned Global Address Space (PGAS) UPC, OpenSHMEM, CAF, UPC++ etc. Hybrid Programming: MPI + PGAS (OpenSHMEM, UPC) Deep Learning Caffe, CNTK, TensorFlow, and many more Big Data/Enterprise/Commercial Computing Focuses on large data and data analysis Spark and Hadoop (HDFS, HBase, MapReduce) Memcached is also used for Web 2. OFA (March 17) 4

5 Parallel Programming Models Overview P1 P2 P3 P1 P2 P3 P1 P2 P3 Shared Memory Memory Memory Memory Logical shared memory Memory Memory Memory Shared Memory Model Distributed Memory Model Partitioned Global Address Space (PGAS) SHMEM, DSM MPI (Message Passing Interface) Global Arrays, UPC, Chapel, X1, CAF, Programming models provide abstract machine models Models can be mapped on different types of systems e.g. Distributed Shared Memory (DSM), MPI within a node, etc. PGAS models and Hybrid MPI+PGAS models are gradually receiving importance OFA (March 17) 5

6 OFA (March 17) 6 Partitioned Global Address Space (PGAS) Models Key features - Simple shared memory abstractions - Light weight one-sided communication - Easier to express irregular communication Different approaches to PGAS - Languages - Libraries Unified Parallel C (UPC) OpenSHMEM Co-Array Fortran (CAF) UPC++ X1 Global Arrays Chapel

7 OFA (March 17) 7 Hybrid (MPI+PGAS) Programming Application sub-kernels can be re-written in MPI/PGAS based on communication characteristics Benefits: Best of Distributed Computing Model Best of Shared Memory Computing Model HPC Application Kernel 1 MPI Kernel 2 PGAS MPI Kernel 3 MPI Kernel N PGAS MPI

8 Overview of the MVAPICH2 Project High Performance open-source MPI Library for InfiniBand, Omni-Path, Ethernet/iWARP, and RDMA over Converged Ethernet (RoCE) MVAPICH (MPI-1), MVAPICH2 (MPI-2.2 and MPI-3.), Started in 21, First version available in 22 MVAPICH2-X (MPI + PGAS), Available since 211 Support for GPGPUs (MVAPICH2-GDR) and MIC (MVAPICH2-MIC), Available since 214 Support for Virtualization (MVAPICH2-Virt), Available since 215 Support for Energy-Awareness (MVAPICH2-EA), Available since 215 Support for InfiniBand Network Analysis and Monitoring (OSU INAM) since 215 Used by more than 2,75 organizations in 83 countries More than 412, (>.4 million) downloads from the OSU site directly Empowering many TOP5 clusters (Nov 16 ranking) 1st, 1,649,6-core (Sunway TaihuLight) at National Supercomputing Center in Wuxi, China 13th, 241,18-core (Pleiades) at NASA 17th, 462,462-core (Stampede) at TACC 4th, 74,52-core (Tsubame 2.5) at Tokyo Institute of Technology Available with software stacks of many vendors and Linux Distros (RedHat and SuSE) Empowering Top5 systems for over a decade System-X from Virginia Tech (3 rd in Nov 23, 2,2 processors, TFlops) -> Sunway TaihuLight (1 st in Jun 16, 1M cores, 1 PFlops) OFA (March 17) 8

9 45 MVAPICH2 Release Timeline and Downloads Number of Downloads MV.9.4 MV2.9. MV2.9.8 MV2 1. MV 1. MV MV 1.1 MV2 1.4 MV2 1.5 MV2 1.6 MV2 1.7 MV2 1.8 MV2 1.9 MV2-GDR 2.b MV2-MIC 2. MV2 2.1 MV2-Virt 2.1rc2 MV2-GDR 2.2rc1 MV2-X 2.2 MV2 2.3a 1 5 Sep-4 Feb-5 Jul-5 Dec-5 May-6 Oct-6 Mar-7 Aug-7 Jan-8 Jun-8 Nov-8 Apr-9 Sep-9 Feb-1 Jul-1 Dec-1 May-11 Oct-11 Mar-12 Aug-12 Jan-13 Jun-13 Timeline OFA (March 17) 9 Nov-13 Apr-14 Sep-14 Feb-15 Jul-15 Dec-15 May-16 Oct-16 Mar-17

10 Architecture of MVAPICH2 Software Family High Performance Parallel Programming Models Message Passing Interface (MPI) PGAS (UPC, OpenSHMEM, CAF, UPC++) Hybrid --- MPI + X (MPI + PGAS + OpenMP/Cilk) High Performance and Scalable Communication Runtime Diverse APIs and Mechanisms Point-topoint Primitives Collectives Algorithms Job Startup Energy- Awareness Remote Memory Access I/O and File Systems Fault Tolerance Virtualization Active Messages Introspection & Analysis Support for Modern Networking Technology (InfiniBand, iwarp, RoCE, Omni-Path) Support for Modern Multi-/Many-core Architectures (Intel-Xeon, OpenPower, Xeon-Phi (MIC, KNL), NVIDIA GPGPU) Transport Protocols RC XRC UD DC UMR ODP Modern Features SR- IOV Multi Rail Transport Mechanisms Shared Memory CMA IVSHMEM Modern Features MCDRAM * NVLink * CAPI * * Upcoming OFA (March 17) 1

11 OFA (March 17) 11 MVAPICH2 Software Family High-Performance Parallel Programming Libraries MVAPICH2 MVAPICH2-X MVAPICH2-GDR MVAPICH2-Virt MVAPICH2-EA MVAPICH2-MIC Microbenchmarks OMB Tools OSU INAM OEMT Support for InfiniBand, Omni-Path, Ethernet/iWARP, and RoCE Advanced MPI features, OSU INAM, PGAS (OpenSHMEM, UPC, UPC++, and CAF), and MPI+PGAS programming models with unified communication runtime Optimized MPI for clusters with NVIDIA GPUs High-performance and scalable MPI for hypervisor and container based HPC cloud Energy aware and High-performance MPI Optimized MPI for clusters with Intel KNC Microbenchmarks suite to evaluate MPI and PGAS (OpenSHMEM, UPC, and UPC++) libraries for CPUs and GPUs Network monitoring, profiling, and analysis for clusters with MPI and scheduler integration Utility to measure the energy consumption of MPI applications

12 OFA (March 17) 12 MVAPICH2 2.3a Released on 3/29/217 Major Features and Enhancements Based on and ABI compatible with MPICH 3.2 Support collective offload using Mellanox's SHArP for Allreduce Enhance tuning framework for Allreduce using SHArP Introduce capability to run MPI jobs across multiple InfiniBand subnets Introduce basic support for executing MPI jobs in Singularity Enhance collective tuning for Intel Knight's Landing and Intel Omni-path Enhance process mapping support for multi-threaded MPI applications Introduce MV2_CPU_BINDING_POLICY=hybrid Introduce MV2_THREADS_PER_PROCESS On-demand connection management for PSM-CH3 and PSM2-CH3 channels Enhance PSM-CH3 and PSM2-CH3 job startup to use non-blocking PMI calls Enhance debugging support for PSM-CH3 and PSM2-CH3 channels Improve performance of architecture detection Introduce run time parameter MV2_SHOW_HCA_BINDING to show process to HCA binding Enhance MV2_SHOW_CPU_BINDING to enable display of CPU bindings on all nodes Deprecate OFA-IB-Nemesis channel Update to hwloc version

13 OFA (March 17) 13 Overview of A Few Challenges being Addressed by the MVAPICH2 Project for Exascale Scalability for million to billion processors Support for highly-efficient inter-node and intra-node communication Support for advanced IB mechanisms (UMR and ODP) Extremely minimal memory footprint (DCT) Scalable Job Start-up Dynamic and Adaptive Tag Matching Collective Support with SHArP Unified Runtime for Hybrid MPI+PGAS programming (MPI + OpenSHMEM, MPI + UPC, CAF, UPC++, ) Integrated Support for GPGPUs Energy-Awareness InfiniBand Network Analysis and Monitoring (INAM) Virtualization and HPC Cloud

14 One-way Latency: MPI over IB with MVAPICH2 Latency (us) 1.8 Small Message Latency Latency (us) Large Message Latency TrueScale-QDR ConnectX-3-FDR ConnectIB-DualFDR ConnectX-4-EDR Omni-Path Message Size (bytes) Message Size (bytes) TrueScale-QDR GHz Deca-core (Haswell) Intel PCI Gen3 with IB switch ConnectX-3-FDR GHz Deca-core (IvyBridge) Intel PCI Gen3 with IB switch ConnectIB-Dual FDR GHz Deca-core (Haswell) Intel PCI Gen3 with IB switch ConnectX-4-EDR GHz Deca-core (Haswell) Intel PCI Gen3 with IB Switch Omni-Path GHz Deca-core (Haswell) Intel PCI Gen3 with Omni-Path switch OFA (March 17) 14

15 Bandwidth (MBytes/sec) Bandwidth: MPI over IB with MVAPICH2 14 Unidirectional Bandwidth 12, , , ,356 3,373 Bandwidth (MBytes/sec) Bidirectional Bandwidth TrueScale-QDR ConnectX-3-FDR ConnectIB-DualFDR ConnectX-4-EDR Omni-Path 24,136 21,983 21,227 12,161 6,228 Message Size (bytes) Message Size (bytes) TrueScale-QDR GHz Deca-core (Haswell) Intel PCI Gen3 with IB switch ConnectX-3-FDR GHz Deca-core (IvyBridge) Intel PCI Gen3 with IB switch ConnectIB-Dual FDR GHz Deca-core (Haswell) Intel PCI Gen3 with IB switch ConnectX-4-EDR GHz Deca-core (Haswell) Intel PCI Gen3 IB switch Omni-Path GHz Deca-core (Haswell) Intel PCI Gen3 with Omni-Path switch OFA (March 17) 15

16 MVAPICH2 Two-Sided Intra-Node Performance (Shared memory and Kernel-based Zero-copy Support (LiMIC and CMA)) Latency (us) 1.5 Latency Intra-Socket Inter-Socket.45 us.18 us Intel Ivy-bridge K Message Size (Bytes) 1 5 Bandwidth (MB/s)15 Bandwidth (Intra-socket) intra-socket-cma intra-socket-shmem intra-socket-limic 14,25 MB/s 1 5 Bandwidth (MB/s)15 Bandwidth (Inter-socket) inter-socket-cma inter-socket-shmem inter-socket-limic 13,749 MB/s Message Size (Bytes) Message Size (Bytes) OFA (March 17) 16

17 User-mode Memory Registration (UMR) Introduced by Mellanox to support direct local and remote noncontiguous memory access Avoid packing at sender and unpacking at receiver Available since MVAPICH2-X 2.2b Latency (us) Small & Medium Message Latency UMR Default 4K 16K 64K 256K 1M 2M 4M 8M 16M Message Size (Bytes) Message Size (Bytes) Connect-IB (54 Gbps): 2.8 GHz Dual Ten-core (IvyBridge) Intel PCI Gen3 with Mellanox IB FDR switch Latency (us) Large Message Latency UMR Default M. Li, H. Subramoni, K. Hamidouche, X. Lu and D. K. Panda, High Performance MPI Datatype Support with User-mode Memory Registration: Challenges, Designs and Benefits, CLUSTER, 215 OFA (March 17) 17

18 Minimizing Memory Footprint by Direct Connect (DC) Transport Node 3 P6 P7 Node P Node 2 P4 P1 IB Network P5 Node 1 P2 P3 Constant connection cost (One QP for any peer) Full Feature Set (RDMA, Atomics etc) Separate objects for send (DC Initiator) and receive (DC Target) DC Target identified by DCT Number Messages routed with (DCT Number, LID) Requires same DC Key to enable communication Available since MVAPICH2-X 2.2a NAMD - Apoa1: Large data set RC DC-Pool UD XRC 97 RC DC-Pool UD XRC Number of Processes Number of Processes H. Subramoni, K. Hamidouche, A. Venkatesh, S. Chakraborty and D. K. Panda, Designing MPI Library with Dynamic Connected Transport (DCT) of InfiniBand : Early Experiences. IEEE International Supercomputing Conference (ISC 14) OFA (March 17) 18 Connection Memory (KB) Memory Footprint for Alltoall Normalized Execution Time

19 Towards High Performance and Scalable Startup at Exascale Memory Required to Store Endpoint Information P P M O a b c M PGAS State of the art MPI State of the art PGAS/MPI Optimized Job Startup Performance d e O a b c d e On-demand Connection PMIX_Ring PMIX_Ibarrier PMIX_Iallgather Shmem based PMI Near-constant MPI and OpenSHMEM initialization time at any process count 1x and 3x improvement in startup time of MPI and OpenSHMEM respectively at 16,384 processes Memory consumption reduced for remote endpoint information by O(processes per node) 1GB Memory saved per node with 1M processes and 16 processes per node a On-demand Connection Management for OpenSHMEM and OpenSHMEM+MPI. S. Chakraborty, H. Subramoni, J. Perkins, A. A. Awan, and D K Panda, 2th International Workshop on High-level Parallel Programming Models and Supportive Environments (HIPS 15) b PMI Extensions for Scalable MPI Startup. S. Chakraborty, H. Subramoni, A. Moody, J. Perkins, M. Arnold, and D K Panda, Proceedings of the 21st European MPI Users' Group Meeting (EuroMPI/Asia 14) c d Non-blocking PMI Extensions for Fast MPI Startup. S. Chakraborty, H. Subramoni, A. Moody, A. Venkatesh, J. Perkins, and D K Panda, 15th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing (CCGrid 15) e SHMEMPMI Shared Memory based PMI for Improved Performance and Scalability. S. Chakraborty, H. Subramoni, J. Perkins, and D K Panda, 16th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing (CCGrid 16) OFA (March 17) 19

20 Time Taken for MPI_Init (seconds) Scalable Job Startup with Non-blocking PMI Co-designing Resource Manager and MPI library K 4K 16K Hello World - MV Default SLURM MPI_Init - MV Default SLURM Hello World - MV Optimized SLURM MPI_Init - MV Optimized SLURM Number of Processes Near-constant MPI_Init at any scale MPI_Init is 59 times faster Hello World (MPI_Init + MPI_Finalize) takes 5.7 times less time with 8,192 processes (512 nodes) 16,384 processes, 1,24 Nodes (Sandy Bridge + FDR) Time Taken (Seconds) 3 Resource Manager Independent Design Hello World MPI_Init K 4K 16K 64K Number of Processes Non-blocking PMI implementation in mpirun_rsh On-demand connection management for PSM and Omni-path Efficient intra-node startup for Knights Landing MPI_Init takes 5.8 seconds Hello World takes 21 seconds 65,536 processes, 1,24 Nodes (KNL + Omni-Path) New designs available in MVAPICH2-2.3a and as patch for SLURM and SLURM OFA (March 17) 2

21 On-Demand Paging (ODP) Introduced by Mellanox to avoid pinning the pages of registered memory regions ODP-aware runtime could reduce the size of pin-down buffers while maintaining performance Available in MVAPICH2-X 2.2 ExecutionTime (s) Pin-down ODP Applications (64 Processes) M. Li, K. Hamidouche, X. Lu, H. Subramoni, J. Zhang, and D. K. Panda, Designing MPI Library with On-Demand Paging (ODP) of InfiniBand: Challenges and Benefits, SC, 216 OFA (March 17) 21

22 Dynamic and Adaptive Tag Matching Challenge Tag matching is a significant overhead for receivers Existing Solutions are - Static and do not adapt dynamically to communication pattern - Do not consider memory overhead Solution A new tag matching design - Dynamically adapt to communication patterns - Use different strategies for different ranks - Decisions are based on the number of request object that must be traversed before hitting on the required one Results Better performance than other state-of-the art tagmatching schemes Minimum memory consumption Will be available in future MVAPICH2 releases Normalized Total Tag Matching Time at 512 Processes Normalized to Default (Lower is Better) Normalized Memory Overhead per Process at 512 Processes Compared to Default (Lower is Better) Adaptive and Dynamic Design for MPI Tag Matching; M. Bayatpour, H. Subramoni, S. Chakraborty, and D. K. Panda; IEEE Cluster 216. [Best Paper Nominee] OFA (March 17) 22

23 Latency (us) Advanced Allreduce Collective Designs Using Switch Offload Mechanism (SHArP) Mesh Refinement Latency (sec) *PPN: Processes Per Node 12 1 MVAPICH2 MVAPICH2-SHArP osu_allreduce (OSU Micro Benchmark) with 448 processes (28 *PPN 16 Nodes) Small-Messages Latency (us) MVAPICH2 MVAPICH2-SHArP Medium-Messages 1K 2K 4K 8K 16K 32K Message Size (Bytes) Message Size (Bytes) Avg of Mesh Refinement Latency for miniamr Application (28 PPN) 224 Processes (8 Nodes, 28 PPN) MVAPICH2 MVAPICH2-SHArP 2.8x 26% 16K 32K 64K Mesh Refinements Levels Mesh Refinement Latency (sec) Processes (16 Nodes, 28 PPN) MVAPICH2 MVAPICH2-SHArP 22% 16K 32K 64K Mesh Refinements Levels 39% Basic Support Available in MVAPICH2 2.3a OFA (March 17) 23

24 OFA (March 17) 24 Overview of A Few Challenges being Addressed by the MVAPICH2 Project for Exascale Scalability for million to billion processors Unified Runtime for Hybrid MPI+PGAS programming (MPI + OpenSHMEM, MPI + UPC, CAF, UPC++, ) Integrated Support for GPGPUs Energy-Awareness InfiniBand Network Analysis and Monitoring (INAM) Virtualization and HPC Cloud

25 MVAPICH2-X for Hybrid MPI + PGAS Applications Current Model Separate Runtimes for OpenSHMEM/UPC/UPC++/CAF and MPI Possible deadlock if both runtimes are not progressed Consumes more network resource Unified communication runtime for MPI, UPC, UPC++, OpenSHMEM, CAF Available with since 212 (starting with MVAPICH2-X 1.9) OFA (March 17) 25

26 UPC++ Support in MVAPICH2-X Time (ms) GASNet_MPI GASNET_IBV MV2-X 1K 2K 4K 8K 16K 32K 64K 128K256K512K 1M 14x MPI + {UPC++} Application UPC++ Runtime GASNet Interfaces Conduit (MPI) MPI Runtime MPI + {UPC++} Application UPC++ Interface MPI Interfaces MVAPICH2-X Unified Communication Runtime (UCR) Message Size (bytes) Network Inter-node Broadcast (64 nodes 1:ppn) Full and native support for hybrid MPI + UPC++ applications Better performance compared to IBV and MPI conduits OSU Micro-benchmarks (OMB) support for UPC++ Available since MVAPICH2-X (2.2rc1) OFA (March 17) 26

27 Application Level Performance with Graph5 and Sort Time (s) Graph5 Execution Time MPI-Simple MPI-CSC MPI-CSR Hybrid (MPI+OpenSHMEM) 4K 8K 16K No. of Processes 7.6X J. Jose, S. Potluri, K. Tomko and D. K. Panda, Designing Scalable Graph5 Benchmark with Hybrid MPI+OpenSHMEM Programming Models, International Supercomputing Conference (ISC 13), June 213 J. Jose, K. Kandalla, M. Luo and D. K. Panda, Supporting Hybrid MPI and OpenSHMEM over InfiniBand: Design and Performance Evaluation, Int'l Conference on Parallel Processing (ICPP '12), September X Performance of Hybrid (MPI+ OpenSHMEM) Graph5 Design 8,192 processes - 2.4X improvement over MPI-CSR - 7.6X improvement over MPI-Simple 16,384 processes - 1.5X improvement over MPI-CSR - 13X improvement over MPI-Simple J. Jose, K. Kandalla, S. Potluri, J. Zhang and D. K. Panda, Optimizing Collective Communication in OpenSHMEM, Int'l Conference on Partitioned Global Address Space Programming Models (PGAS '13), October 213. Time (seconds) MPI Sort Execution Time Hybrid 5GB-512 1TB-1K 2TB-2K 4TB-4K Input Data - No. of Processes Performance of Hybrid (MPI+OpenSHMEM) Sort Application 4,96 processes, 4 TB Input Size - MPI 248 sec;.16 TB/min - Hybrid 1172 sec;.36 TB/min - 51% improvement over MPI-design OFA (March 17) 27 51%

28 OFA (March 17) 28 Overview of A Few Challenges being Addressed by the MVAPICH2 Project for Exascale Scalability for million to billion processors Unified Runtime for Hybrid MPI+PGAS programming (MPI + OpenSHMEM, MPI + UPC, CAF, UPC++, ) Integrated Support for GPGPUs CUDA-aware MPI GPUDirect RDMA (GDR) Support Energy-Awareness InfiniBand Network Analysis and Monitoring (INAM) Virtualization and HPC Cloud

29 MPI + CUDA - Naive Data movement in applications with standard MPI and CUDA interfaces At Sender: CPU cudamemcpy(s_hostbuf, s_devbuf,...); MPI_Send(s_hostbuf, size,...); At Receiver: MPI_Recv(r_hostbuf, size,...); cudamemcpy(r_devbuf, r_hostbuf,...); PCIe GPU NIC Switch High Productivity and Low Performance OFA (March 17) 29

30 MPI + CUDA - Advanced Pipelining at user level with non-blocking MPI and CUDA interfaces At Sender: for (j = ; j < pipeline_len; j++) cudamemcpyasync(s_hostbuf + j * blk, s_devbuf + j * blksz, ); for (j = ; j < pipeline_len; j++) { while (result!= cudasucess) { result = cudastreamquery( ); if(j > ) MPI_Test( ); } MPI_Isend(s_hostbuf + j * block_sz, blksz...); } MPI_Waitall(); <<Similar at receiver>> Low Productivity and High Performance CPU PCIe GPU NIC Switch OFA (March 17) 3

31 GPU-Aware (CUDA-Aware) MPI Library: MVAPICH2-GPU Standard MPI interfaces used for unified data movement Takes advantage of Unified Virtual Addressing (>= CUDA 4.) Overlaps data movement from GPU with RDMA transfers At Sender: MPI_Send(s_devbuf, size, ); At Receiver: MPI_Recv(r_devbuf, size, ); inside MVAPICH2 High Performance and High Productivity OFA (March 17) 31

32 CUDA-Aware MPI: MVAPICH2-GDR Releases Support for MPI communication from NVIDIA GPU device memory High performance RDMA-based inter-node point-to-point communication (GPU-GPU, GPU-Host and Host-GPU) High performance intra-node point-to-point communication for multi-gpu adapters/node (GPU-GPU, GPU-Host and Host-GPU) Taking advantage of CUDA IPC (available since CUDA 4.1) in intra-node communication for multiple GPU adapters/node Optimized and tuned collectives for GPU device buffers MPI datatype support for point-to-point and collective communication from GPU device buffers Unified memory OFA (March 17) 32

33 Latency (us) Performance of MVAPICH2-GPU with GPU-Direct RDMA (GDR) Bi-Bandwidth (MB/s) us K Message Size (bytes) GPU-GPU Internode Bi-Bandwidth MV2-GDR2.2 MV2-GDR2.b MV2 w/o GDR K 4K GPU-GPU internode latency MV2-GDR2.2 MV2-GDR2.b MV2 w/o GDR 3X 1x Message Size (bytes) 11x Bandwidth (MB/s) 2X GPU-GPU Internode Bandwidth MV2-GDR2.2 MV2-GDR2.b MV2 w/o GDR K 4K Message Size (bytes) 11X MVAPICH2-GDR-2.2 Intel Ivy Bridge (E5-268 v2) node - 2 cores NVIDIA Tesla K4c GPU Mellanox Connect-X4 EDR HCA CUDA 8. Mellanox OFED 3. with GPU-Direct-RDMA OFA (March 17) 33 2X

34 Average Time Steps per second (TPS) Application-Level Evaluation (HOOMD-blue) K Particles Platform: Wilkes (Intel Ivy Bridge + NVIDIA Tesla K2c + Mellanox Connect-IB) HoomdBlue Version 1..5 GDRCOPY enabled: MV2_USE_CUDA=1 MV2_IBA_HCA=mlx5_ MV2_IBA_EAGER_THRESHOLD=32768 MV2_VBUF_TOTAL_SIZE=32768 MV2_USE_GPUDIRECT_LOOPBACK_LIMIT=32768 MV2_USE_GPUDIRECT_GDRCOPY=1 MV2_USE_GPUDIRECT_GDRCOPY_LIMIT= X Number of Processes Average Time Steps per second (TPS) K Particles MV2 MV2+GDR Number of Processes OFA (March 17) 34 2X

35 Application-Level Evaluation (Cosmo) and Weather Forecasting in Switzerland Normalized Execution Time Wilkes GPU Cluster Default Callback-based Event-based Number of GPUs Normalized Execution Time CSCS GPU cluster Default Callback-based Event-based Number of GPUs 2X improvement on 32 GPUs nodes 3% improvement on 96 GPU nodes (8 GPUs/node) Cosmo model: /tasks/operational/meteoswiss/ On-going collaboration with CSCS and MeteoSwiss (Switzerland) in co-designing MV2-GDR and Cosmo Application C. Chu, K. Hamidouche, A. Venkatesh, D. Banerjee, H. Subramoni, and D. K. Panda, Exploiting Maximal Overlap for Non-Contiguous Data Movement Processing on Modern GPU-enabled Systems, IPDPS 16 OFA (March 17) 35

36 OFA (March 17) 36 Overview of A Few Challenges being Addressed by the MVAPICH2 Project for Exascale Scalability for million to billion processors Unified Runtime for Hybrid MPI+PGAS programming (MPI + OpenSHMEM, MPI + UPC, CAF, UPC++, ) Integrated Support for GPGPUs Energy-Awareness InfiniBand Network Analysis and Monitoring (INAM)

37 OFA (March 17) 37 Energy-Aware MVAPICH2 & OSU Energy Management Tool (OEMT) MVAPICH2-EA 2.1 (Energy-Aware) A white-box approach New Energy-Efficient communication protocols for pt-pt and collective operations Intelligently apply the appropriate Energy saving techniques Application oblivious energy saving OEMT A library utility to measure energy consumption for MPI applications Works with all MPI runtimes PRELOAD option for precompiled applications Does not require ROOT permission: A safe kernel module to read only a subset of MSRs Publicly available since August 15

38 OFA (March 17) 38 MVAPICH2-EA: Application Oblivious Energy-Aware-MPI (EAM) An energy efficient runtime that provides energy savings without application knowledge Uses automatically and transparently the best energy lever Provides guarantees on maximum degradation with 5-41% savings at <= 5% degradation Pessimistic MPI applies energy reduction lever to each MPI call 1 A Case for Application-Oblivious Energy-Efficient MPI Runtime A. Venkatesh, A. Vishnu, K. Hamidouche, N. Tallent, D. K. Panda, D. Kerbyson, and A. Hoise, Supercomputing 15, Nov 215 [Best Student Paper Finalist]

39 OFA (March 17) 39 Overview of A Few Challenges being Addressed by the MVAPICH2 Project for Exascale Scalability for million to billion processors Unified Runtime for Hybrid MPI+PGAS programming (MPI + OpenSHMEM, MPI + UPC, CAF, UPC++, ) Integrated Support for GPGPUs Energy-Awareness InfiniBand Network Analysis and Monitoring (INAM)

40 OFA (March 17) 4 Overview of OSU INAM A network monitoring and analysis tool that is capable of analyzing traffic on the InfiniBand network with inputs from the MPI runtime Monitors IB clusters in real time by querying various subnet management entities and gathering input from the MPI runtimes OSU INAM v.9.1 released on 5/13/16 Significant enhancements to user interface to enable scaling to clusters with thousands of nodes Improve database insert times by using 'bulk inserts Capability to look up list of nodes communicating through a network link Capability to classify data flowing over a network link at job level and process level granularity in conjunction with MVAPICH2-X 2.2rc1 Best practices guidelines for deploying OSU INAM on different clusters Capability to analyze and profile node-level, job-level and process-level activities for MPI communication Point-to-Point, Collectives and RMA Ability to filter data based on type of counters using drop down list Remotely monitor various metrics of MPI processes at user specified granularity "Job Page" to display jobs in ascending/descending order of various performance metrics in conjunction with MVAPICH2-X Visualize the data transfer happening in a live or historical fashion for entire network, job or set of nodes

41 OSU INAM Features --- Clustered View (1,879 nodes, 212 switches, 4,377 network links) Finding Routes Between Nodes Show network topology of large clusters Visualize traffic pattern on different links Quickly identify congested links/links in error state See the history unfold play back historical state of the network OFA (March 17) 41

42 OSU INAM Features (Cont.) Job level view Visualizing a Job (5 Nodes) Show different network metrics (load, error, etc.) for any live job Play back historical data for completed jobs to identify bottlenecks Node level view - details per process or per node CPU utilization for each rank/node Bytes sent/received for MPI operations (pt-to-pt, collective, RMA) Network metrics (e.g. XmitDiscard, RcvError) per rank/node Estimated Process Level Link Utilization Estimated Link Utilization view Classify data flowing over a network link at different granularity in conjunction with MVAPICH2-X 2.2rc1 Job level and Process level OFA (March 17) 42

43 Major Computing Categories Scientific Computing Message Passing Interface (MPI), including MPI + OpenMP, is the Dominant Programming Model Many discussions towards Partitioned Global Address Space (PGAS) UPC, OpenSHMEM, CAF, UPC++ etc. Hybrid Programming: MPI + PGAS (OpenSHMEM, UPC) Deep Learning Caffe, CNTK, TensorFlow, and many more OFA (March 17) 43

44 Deep Learning and MPI: State-of-the-art Deep Learning is going through a resurgence Excellent accuracy for deep/convolutional neural networks Public availability of versatile datasets like MNIST, CIFAR, and ImageNet Widespread popularity of accelerators like Nvidia GPUs DL frameworks and applications Caffe, Microsoft CNTK, Google TensorFlow and many more.. Most of the frameworks are exploiting GPUs to accelerate training Diverse range of applications Image Recognition, Cancer Detection, Self-Driving Cars, Speech Processing etc. Can MPI runtimes like MVAPICH2 provide efficient support for Deep Learning workloads? MPI runtimes typically deal with relatively small-sizes message (order of kilobytes) CPU-based communication buffers Relative Search Interest Deep Learning - Google Trends Years Deep Learning - Google Trends OFA (March 17) 44

45 Deep Learning: New Challenges for MPI Runtimes Deep Learning frameworks are a different game altogether Unusually large message sizes (order of megabytes) Most communication based on GPU buffers How to address these newer requirements? GPU-specific Communication Libraries (NCCL) NVidia's NCCL library provides inter-gpu communication CUDA-Aware MPI (MVAPICH2-GDR) Provides support for GPU-based communication Can we exploit CUDA-Aware MPI and NCCL to support Deep Learning applications? Hierarchical Communication (Knomial + NCCL ring) OFA (March 17) 45

46 Efficient Broadcast: MVAPICH2-GDR and NCCL NCCL has some limitations Only works for a single node, thus, no scale-out on multiple nodes Degradation across IOH (socket) for scale-up (within a node) We propose optimized MPI_Bcast Communication of very large GPU buffers (order of megabytes) Scale-out on large number of dense multi-gpu nodes Hierarchical Communication that efficiently exploits: K 32K 256K 2M 16M 128M 2 CUDA-Aware MPI_Bcast in MV2-GDR 1 NCCL Broadcast primitive Efficient Large Message Broadcast using NCCL and CUDA-Aware MPI for Deep Learning, A. Awan, K. Hamidouche, A. Venkatesh, and D. K. Panda, Number of GPUs The 23rd European MPI Users' Group Meeting (EuroMPI 16), Sep 216 [Best Paper Runner-Up] Performance Benefits: Microsoft CNTK DL framework (25% avg. improvement ) OFA (March 17) 46 Latency (us) Log Scale Time (seconds) 3 MV2-GDR Message Size MV2-GDR MV2-GDR-Opt Performance Benefits: OSU Micro-benchmarks MV2-GDR-Opt 1x

47 Large Message Optimized Collectives for Deep Learning MV2-GDR provides optimized collectives for large message sizes Optimized Reduce, Allreduce, and Bcast Good scaling with large number of GPUs Available with MVAPICH2- GDR 2.2GA Latency (ms) Latency (ms) Latency (ms) Reduce 192 GPUs Message Size (MB) Allreduce 64 GPUs Message Size (MB) Bcast 64 GPUs Message Size (MB) No. of GPUs OFA (March 17) 47 Latency (ms) Latency (ms) Latency (ms) Reduce 64 MB No. of GPUs Allreduce MB No. of GPUs Bcast 128 MB

48 OSU-Caffe: Scalable Deep Learning Caffe : A flexible and layered Deep Learning framework. Benefits and Weaknesses Multi-GPU Training within a single node Performance degradation for GPUs across different sockets Limited Scale-out OSU-Caffe: MPI-based Parallel Training Enable Scale-up (within a node) and Scale-out (across multi-gpu nodes) network on ImageNet dataset Awan, K. Hamidouche, J. Hashmi, and D. K. Panda, S-Caffe: Co-designing MPI Runtimes and Caffe for Scalable Deep Learning on Modern GPU Clusters, PPoPP, Sep 217 OSU-Caffe is publicly available from: Training Time (seconds) GoogLeNet (ImageNet) on 128 GPUs 5 Invalid use case No. of GPUs Caffe OSU-Caffe (124) OSU-Caffe (248) OFA (March 17) 48

49 OFA (March 17) 49 MVAPICH2 Plans for Exascale Performance and Memory scalability toward 1M cores Hybrid programming (MPI + OpenSHMEM, MPI + UPC, MPI + CAF ) MPI + Task* Enhanced Optimization for GPU Support and Accelerators Enhanced communication schemes for upcoming architectures Knights Landing with MCDRAM* NVLINK* CAPI* Extended topology-aware collectives Extended Energy-aware designs and Virtualization Support Extended Support for MPI Tools Interface (as in MPI 3.1) Extended Checkpoint-Restart and migration support with SCR Support for * features will be available in future MVAPICH2 Releases

50 OFA (March 17) 5 Two More Presentations Thursday (3/3/17) at 9:am Building Efficient HPC Clouds with MVAPICH2 and RDMA-Hadoop over SR-IOV IB Clusters Friday (3/31/17) at 11:am NVM-aware RDMA-Based Communication and I/O Schemes for High-Perf Big Data Analytics

51 OFA (March 17) 51 Funding Acknowledgments Funding Support by Equipment Support by

52 Personnel Acknowledgments Current Students A. Awan (Ph.D.) R. Biswas (M.S.) M. Bayatpour (Ph.D.) S. Chakraborthy (Ph.D.) C.-H. Chu (Ph.D.) S. Guganani (Ph.D.) J. Hashmi (Ph.D.) H. Javed (Ph.D.) M. Li (Ph.D.) D. Shankar (Ph.D.) H. Shi (Ph.D.) J. Zhang (Ph.D.) Current Research Scientists X. Lu H. Subramoni Current Research Specialist J. Smith Past Students A. Augustine (M.S.) P. Balaji (Ph.D.) S. Bhagvat (M.S.) A. Bhat (M.S.) D. Buntinas (Ph.D.) L. Chai (Ph.D.) B. Chandrasekharan (M.S.) N. Dandapanthula (M.S.) V. Dhanraj (M.S.) T. Gangadharappa (M.S.) K. Gopalakrishnan (M.S.) W. Huang (Ph.D.) W. Jiang (M.S.) J. Jose (Ph.D.) S. Kini (M.S.) M. Koop (Ph.D.) K. Kulkarni (M.S.) R. Kumar (M.S.) S. Krishnamoorthy (M.S.) K. Kandalla (Ph.D.) P. Lai (M.S.) J. Liu (Ph.D.) M. Luo (Ph.D.) A. Mamidala (Ph.D.) G. Marsh (M.S.) V. Meshram (M.S.) A. Moody (M.S.) S. Naravula (Ph.D.) R. Noronha (Ph.D.) X. Ouyang (Ph.D.) S. Pai (M.S.) S. Potluri (Ph.D.) R. Rajachandrasekar (Ph.D.) G. Santhanaraman (Ph.D.) A. Singh (Ph.D.) J. Sridhar (M.S.) S. Sur (Ph.D.) H. Subramoni (Ph.D.) K. Vaidyanathan (Ph.D.) A. Vishnu (Ph.D.) J. Wu (Ph.D.) W. Yu (Ph.D.) Past Research Scientist K. Hamidouche S. Sur Past Programmers D. Bureddy M. Arnold J. Perkins Past Post-Docs D. Banerjee X. Besseron H.-W. Jin J. Lin M. Luo E. Mancini S. Marcarelli J. Vienne H. Wang OFA (March 17) 52

53 OFA (March 17) 53 Thank You! Network-Based Computing Laboratory The MVAPICH2 Project The High-Performance Deep Learning Project