Building a GPU-enabled OpenStack Cloud for HPC - Blair Bethwaite, Monash University
1 like461 views
Audience Level Intermediate Synopsis M3 is the latest generation system of the MASSIVE project, an HPC facility specializing in characterization science (imaging and visualization). Using OpenStack as the compute provisioning layer, M3 is a hybrid HPC/cloud system, custom-integrated by Monash’s R@CMon Research Cloud team. Built to support Monash University’s next-gen high-throughput instrument processing requirements, M3 is half-half GPU-accelerated and CPU-only. We’ll discuss the design and tech used to build this innovative platform as well as detailing approaches and challenges to building GPU-enabled and HPC clouds. We’ll also discuss some of the software and processing pipelines that this system supports and highlight the importance of tuning for these workloads. Speaker Bio Blair Bethwaite: Blair has worked in distributed computing at Monash University for 10 years, with OpenStack for half of that. Having served as team lead, architect, administrator, user, researcher, and occasional hacker, Blair’s unique perspective as a science power-user, developer, and system architect has helped guide the evolution of the research computing engine central to Monash’s 21st Century Microscope. Lance Wilson: Lance is a mechanical engineer, who has been making tools to break things for the last 20 years. His career has moved through a number of engineering subdisciplines from manufacturing to bioengineering. Now he supports the national characterisation research community in Melbourne, Australia using OpenStack to create HPC systems solving problems too large for your laptop.
![bought to you by
MASSIVE Business Plan 2013 / 2014 DRAFT
GPU-accelerated OpenStack Instances
1. Confirm hardware capability
2. Prep compute hosts/hypervisors
1. ensure IOMMU is enabled in BIOS
2. enable IOMMU in Linux, e.g., for Intel:
3. ensure no other drivers/modules claim GPUs, e.g., blacklist
nouveau
4. Configure nova-compute.conf pci_passthrough_whitelist:
# in /etc/default/grub:
GRUB_CMDLINE_LINUX_DEFAULT="quiet splash intel_iommu=on iommu=pt rd.modules-
load=vfio-pci”
~$ update-grub
~$ lspci -nn | grep NVIDIA
03:00.0 3D controller [0302]: NVIDIA Corporation Device [10de:15f8] (rev a1)
82:00.0 3D controller [0302]: NVIDIA Corporation Device [10de:15f8] (rev a1)
# in /etc/nova/nova.conf:
pci_passthrough_whitelist=[{"vendor_id":"10de", "product_id":"15f8"}]](https://image.slidesharecdn.com/openstack-gpu-hpc-openstackdaymelbourneblairmonash-190624050233/85/Building-a-GPU-enabled-OpenStack-Cloud-for-HPC-Blair-Bethwaite-Monash-University-20-320.jpg)
![[OpenStack Day in Korea 2015] Track 2-3 - 오픈스택 클라우드에 최적화된 네트워크 가상화 '누아지(Nuage)'](https://cdn.slidesharecdn.com/ss_thumbnails/23-150213052422-conversion-gate01-thumbnail.jpg?width=560&fit=bounds)
Ad
More Related Content
What's hot (20)
Similar to Building a GPU-enabled OpenStack Cloud for HPC - Blair Bethwaite, Monash University (20)
Ad
More from OpenStack (20)
Ad
Recently uploaded (20)
Building a GPU-enabled OpenStack Cloud for HPC - Blair Bethwaite, Monash University
- 1. Building a GPU-enabled OpenStack Cloud for HPC Blair Bethwaite, Lance Wilson, (and many others) MONASH eRESEARCH
- 2. Monash eResearch Centre: Enabling and Accelerating 21st Century Discovery through the application of advanced computing, data informatics, tools and infrastructure, delivered at scale, and built by with “co-design” principle (researcher + technologist)
- 3. bought to you by MASSIVE Business Plan 2013 / 2014 DRAFT • UniMelb, as lead agent for Nectar, established first Node/site of the Research Cloud in Jan 2012 and opened doors to the research community • Now seven Nodes (10+ DCs) and >40k cores around Australia • Nectar established an OpenStack ecosystem for research computing in Australia • M3 built as first service in a new “monash-03” zone of the Research Cloud focusing on HPC (computing) & HPDA (data-analytics)
- 4. bought to you by MASSIVE Business Plan 2013 / 2014 DRAFT
- 5. HPC 150 active projects 1000+ user accounts 100+ institutions across Australia Interactive Vis 600+ users Multi-modal Australian ScienceS Imaging and Visualisation Environment Specialised Facility for Imaging and Visualisation MASSIVE Instrument Integration Integrating with key Australian Instrument Facilities. – IMBL, XFM – CryoEM – MBI – NCRIS: NIF, AMMRF Large cohort of researchers new to HPC ~$2M per year funded by partners and national project funding Partners Monash University Australian Synchrotron CSIRO Affiliate Partners ARC Centre of Excellence in Integrative Brain Function ARC Centre of Excellence in Advanced Molecular Imaging
- 6. M3 at Monash University (including recent upgrade) A Computer for Next-Generation Data Science 2100 Intel Haswell CPU-cores 560 Intel Broadwell CPU-cores NVIDIA GPU coprocessors for data processing and visualisation: • 48 NVIDIA Tesla K80 • 40 NVIDIA Pascal P100 (16GB PCIe) (upgrade) • 8 NVIDIA Grid K1 (32 individual GPUs) for medium and low end visualisation A 1.15 petabyte Lustre parallel file system 100 Gb/s Ethernet Mellanox Spectrum Supplied by Dell, Mellanox and NVIDIA M3 Steve Oberlin, Chief Technology Officer Accelerated Computing, NVIDIA Alan Finkel Australia’s Chief Scientist
- 7. bought to you by MASSIVE Business Plan 2013 / 2014 DRAFT www.openstack.org/science openstack.org The Crossroads of Cloud and HPC: OpenStack for Scientific Research Exploring OpenStack cloud computing for scientific workloads
- 8. bought to you by MASSIVE Business Plan 2013 / 2014 DRAFT Why OpenStack ‣Heterogeneous user requirements ‣same underlying infrastructure can be expanded to accommodate multiple distinct and dynamic clusters (e.g. bioinformatics focused, Hadoop) ‣Clusters need provisioning systems anyway ‣Forcing the cluster to be cloud-provisioning and managed makes it easier to leverage other cloud resources e.g. community science cloud, commercial cloud ‣OpenStack is a big focus of innovation and effort in the industry - benefits of association and osmosis ‣Business function boundaries at the APIs ?
- 9. bought to you by MASSIVE Business Plan 2013 / 2014 DRAFT Key tuning for HPC ‣ With hardware features & software tuning this is very much possible and performance is almost native ‣ CPU host-model / host-passthrough ‣ Expose host CPU and NUMA cell topology ‣ Pin virtual cores to physical cores ‣ Pin virtual memory to physical memory ‣ Back guest memory with hugepages ‣ Disable kernel consolidation features ‣ Remove host network overheads for high- performance data http://frankdenneman.nl/2015/02/27/memory-deep-dive-numa-data-locality/
- 10. bought to you by MASSIVE Business Plan 2013 / 2014 DRAFT https://www.mellanox.com/related-docs/whitepapers/WP_Solving_IO_Bottlenecks.pdf
- 11. bought to you by MASSIVE Business Plan 2013 / 2014 DRAFT M3 HPFS Integration • special flavors for cluster instances which specify a PCI passthrough SRIOV vNIC • hypervisor has NICs with VFs tied to data VLAN(s) • data VLAN is RDMA capable so e.g. Lustre can use o2ib LNET driver
- 12. bought to you by MASSIVE Business Plan 2013 / 2014 DRAFT HPC-Cloud Interconnect …
- 13. bought to you by MASSIVE Business Plan 2013 / 2014 DRAFT M3 Compute Performance Snapshot • Early system and virtualisation tuning on an m3d node • Hardware & hypervisor: • Dell R730, 2x E5-2680 v3 (2x 12 cores, HT off), 256GB RAM, 2x NVIDIA K80 cards, Mellanox CX-4 50GbE DP • Ubuntu Trusty host with Xenial kernel (4.4) and Mitaka Ubuntu Cloud archive hypervisor (QEMU 2.5 + KVM) • (Kernel samepage merging and transparent huge pages disabled to avoid performance noise) • Guest: • M3 large GPU compute flavor (m3d) - 24 cores, 240GB RAM, 4x K80 GPUs, 1x Mellanox CX-4 Virtual Function • CentOS7 guest (3.10 kernel) running High Performance Linpack and Intel Optimised Linpack
- 14. bought to you by MASSIVE Business Plan 2013 / 2014 DRAFT
- 15. bought to you by MASSIVE Business Plan 2013 / 2014 DRAFT
- 16. bought to you by MASSIVE Business Plan 2013 / 2014 DRAFT So, all 👍 ? • Early user on-boarding hit some speed bumps with inconsistent to poor performance on particular codes/ workloads, e.g., slower than legacy clusters • Initial tuning did not include hugepages because… • Couldn’t start 240GB RAM guests backed by static hugepages - initial memory allocation in KVM is single threaded and takes longer than 30 secs after which libvirt gives up and shoots guest • enabled transparent hugepages (THP) for large memory guests, configured 1G static hugepages for everything else and repeated tests for all hosts to ensure no “bad” nodes • benchmarks from m3a nodes: • Dell C6320, 2x E5-2680 v3 (2x 12 cores, HT off), 128GB RAM, Mellanox CX-4 Lx 25GbE DP • Ubuntu Trusty host with Xenial kernel (4.4) and Mitaka Ubuntu Cloud archive hypervisor (QEMU 2.5 + KVM) • M3 standard compute flavor (m3a) - 24 cores, 120GB RAM, 1x Mellanox CX-4 Lx Virtual Function • CentOS7 guest (3.10 kernel) running High Performance Linpack
- 17. 17 500 550 600 650 700 750 0 20,000 40,000 60,000 80,000 100,000 120,000 140,000 Gigaflops Linpack Matrix Size Hypervisor Guest Without Hpages Guest With Hpages m3a nodes High Performance Linpack (HPL) performance characterisation ?
- 19. bought to you by MASSIVE Business Plan 2013 / 2014 DRAFT GPU-accelerated OpenStack Instances How-to? 1. Confirm hardware capability • IOMMU - Intel VT-d, AMD-Vi (common in contemporary servers) • GPU support • https://etherpad.openstack.org/p/GPU-passthrough-model- success-failure 2. Prep nova-compute hosts/hypervisors 3. Configure OpenStack nova-scheduler 4. Create GPU flavor
- 20. bought to you by MASSIVE Business Plan 2013 / 2014 DRAFT GPU-accelerated OpenStack Instances 1. Confirm hardware capability 2. Prep compute hosts/hypervisors 1. ensure IOMMU is enabled in BIOS 2. enable IOMMU in Linux, e.g., for Intel: 3. ensure no other drivers/modules claim GPUs, e.g., blacklist nouveau 4. Configure nova-compute.conf pci_passthrough_whitelist: # in /etc/default/grub: GRUB_CMDLINE_LINUX_DEFAULT="quiet splash intel_iommu=on iommu=pt rd.modules- load=vfio-pci” ~$ update-grub ~$ lspci -nn | grep NVIDIA 03:00.0 3D controller [0302]: NVIDIA Corporation Device [10de:15f8] (rev a1) 82:00.0 3D controller [0302]: NVIDIA Corporation Device [10de:15f8] (rev a1) # in /etc/nova/nova.conf: pci_passthrough_whitelist=[{"vendor_id":"10de", "product_id":"15f8"}]
- 21. bought to you by MASSIVE Business Plan 2013 / 2014 DRAFT GPU-accelerated OpenStack Instances 1. Confirm hardware capability 2. Prep compute hosts/hypervisors 3. Configure OpenStack nova-scheduler 1. On nova-scheduler / cloud-controllers # in /etc/nova/nova.conf: pci_alias={"vendor_id":"10de", "product_id":"15f8", "name":"P100"} scheduler_driver=nova.scheduler.filter_scheduler.FilterScheduler scheduler_available_filters=nova.scheduler.filters.all_filters scheduler_available_filters=nova.scheduler.filters.pci_passthrough_filter .PciPassthroughFilter scheduler_default_filters=RamFilter,ComputeFilter,AvailabilityZoneFilter, ComputeCapabilitiesFilter,ImagePropertiesFilter,PciPassthroughFilter
- 22. bought to you by MASSIVE Business Plan 2013 / 2014 DRAFT GPU-accelerated OpenStack Instances 1. Confirm hardware capability 2. Prep compute hosts/hypervisors 3. Configure OpenStack nova-scheduler 4. Create GPU flavor ~$ openstack flavor create --ram 122880 --disk 30 --vcpus 24 mon.m3.c24r120.2gpu-p100.mlx ~$ openstack flavor set mon.m3.c24r120.2gpu-p100.mlx --property pci_passthrough:alias='P100:2'
- 23. bought to you by MASSIVE Business Plan 2013 / 2014 DRAFT GPU-accelerated OpenStack Instances ~$ openstack flavor show 56cd053c-b6a2-4103-b870-a83dd5d27ec1 +----------------------------+--------------------------------------------+ | Field | Value | +----------------------------+--------------------------------------------+ | OS-FLV-DISABLED:disabled | False | | OS-FLV-EXT-DATA:ephemeral | 1000 | | disk | 30 | | id | 56cd053c-b6a2-4103-b870-a83dd5d27ec1 | | name | mon.m3.c24r120.2gpu-p100.mlx | | os-flavor-access:is_public | False | | properties | pci_passthrough:alias='P100:2,MlxCX4-VF:1' | | ram | 122880 | | rxtx_factor | 1.0 | | swap | | | vcpus | 24 | +----------------------------+--------------------------------------------+ ~$ openstack server list --all-projects --project d99… --flavor 56c… +--------------------------------------+------------+--------+----------------------------------+ | ID | Name | Status | Networks | +--------------------------------------+------------+--------+----------------------------------+ | 1d77bf12-0099-4580-bf6f-36c42225f2c0 | massive003 | ACTIVE | monash-03-internal=10.16.201.20 | +--------------------------------------+------------+--------+----------------------------------+
- 24. bought to you by MASSIVE Business Plan 2013 / 2014 DRAFT GPU Instances - rough edges • Hardware monitoring • No OOB interface to monitor GPU hardware when it is assigned to an instance (and doing so would require loading drivers in the host) • P2P (peer-to-peer multi-GPU) • PCIe topology not available in default guest configuration (not even a PCIe bus on legacy QEMU i440fx machine type) • PCIe ACS (Access Control Services - forces transactions through the Root Complex which blocks/disallows P2P for security)
- 25. bought to you by MASSIVE Business Plan 2013 / 2014 DRAFT GPU Instances - rough edges • PCIe security • Compromised device could access privileged host memory via PCIe ATS (Address Translation Services) • Some special device registers should be blocked/proxied in multi- tenant environment • Common to use cloud images for base OS+driver versioning and standardisation, but new NVIDIA driver versions do not support some existing hardware (e.g. K1) • Requires multiple images or automated driver deployment/config - no big thing just inconvenient
- 26. bought to you by MASSIVE Business Plan 2013 / 2014 DRAFT OpenStack Cyborg - accelerator management … aims to provide a general purpose management framework for acceleration resources (i.e. various types of accelerators such as Crypto cards, GPUs, FPGAs, NVMe/NOF SSDs, ODP, DPDK/SPDK and so on) (https://wiki.openstack.org/wiki/Cyborg) https://review.openstack.org/#/c/448228/
- 27. Title: Business Plan for the Multi-modal Australian ScienceS Imaging and Visualisation Environment (MASSIVE) 2013 / 2014 Document no: MASSIVE-BP-2.3 DRAFT Date: June 2013 Prepared by: Name: Wojtek J Goscinski Title: MASSIVE Coordinator Approved by: Name: MASSIVE Steering Committee Date: Open IaaS: Technology: 30/10/2015 1:59 pmMyTardis | Automatically stores your instrument data for sharing. Menu MyTardis Tech Group Meeting #3 Posted on August 20, 2015August 20, 2015 by steve.androulakissteve.androulakis It’s been months since the last one, so a wealth of activity to report on. MyTardis Automatically stores your instrument data for sharing. Application layers: