Local and Remote byte-addressable NVDIMM High-level Use Cases

Slides:



Advertisements
Similar presentations
Multiple Processor Systems
Advertisements

High Performance Cluster Computing Architectures and Systems Hai Jin Internet and Cluster Computing Center.
SALSA HPC Group School of Informatics and Computing Indiana University.
Serverless Network File Systems. Network File Systems Allow sharing among independent file systems in a transparent manner Mounting a remote directory.
12/02/14 Chet Douglas, DCG Crystal Ridge PE SW Architecture
The Next Evolution of PVFS. Brought Us…. Modular Architecture Support for Standard out of the box kernels Distributed Meta-data Scalability Direct MPI.
File Management Systems
The SNIA NVM Programming Model
CON Software-Defined Networking in a Hybrid, Open Data Center Krishna Srinivasan Senior Principal Product Strategy Manager Oracle Virtual Networking.
1 I/O Management in Representative Operating Systems.
Module 14: Scalability and High Availability. Overview Key high availability features available in Oracle and SQL Server Key scalability features available.
Microsoft Load Balancing and Clustering. Outline Introduction Load balancing Clustering.
RAID-x: A New Distributed Disk Array for I/O-Centric Cluster Computing Kai Hwang, Hai Jin, and Roy Ho.
MCS: Enterprise Communications CoE Architect.
File Systems and N/W attached storage (NAS) | VTU NOTES | QUESTION PAPERS | NEWS | VTU RESULTS | FORUM | BOOKSPAR ANDROID APP.
Infiniband enables scalable Real Application Clusters – Update Spring 2008 Sumanta Chatterjee, Oracle Richard Frank, Oracle.
Storage Systems in HPC John A. Chandy Department of Electrical and Computer Engineering University of Connecticut.
© 2013 Mellanox Technologies 1 NoSQL DB Benchmarking with high performance Networking solutions WBDB, Xian, July 2013.
Distributed Data Stores – Facebook Presented by Ben Gooding University of Arkansas – April 21, 2015.
Microsoft Consultantancy Services Enterprise Communications Global Practice Solutions Architect.
1 Advanced Storage Technologies for High Performance Computing Sorin, Faibish EMC NAS Senior Technologist IDC HPC User Forum, April 14-16, Norfolk, VA.
Institute of Computer and Communication Network Engineering OFC/NFOEC, 6-10 March 2011, Los Angeles, CA Lessons Learned From Implementing a Path Computation.
Multiple Processor Systems. Multiprocessor Systems Continuous need for faster and powerful computers –shared memory model ( access nsec) –message passing.
Distributed Shared Memory: A Survey of Issues and Algorithms B,. Nitzberg and V. Lo University of Oregon.
Module – 4 Intelligent storage system
GigaSpaces Global HTTP Session Sharing October 2013 Massive Web Application Scaling.
Hadoop Hardware Infrastructure considerations ©2013 OpalSoft Big Data.
1 Moshe Shadmon ScaleDB Scaling MySQL in the Cloud.
Module 10: Maintaining High-Availability. Overview Introduction to Availability Increasing Availability Using Failover Clustering Standby Servers and.
Swapping to Remote Memory over InfiniBand: An Approach using a High Performance Network Block Device Shuang LiangRanjit NoronhaDhabaleswar K. Panda IEEE.
Data in the Cloud – I Parallel Databases The Google File System Parallel File Systems.
DNS Dynamic Update Performance Study The Purpose Dynamic update and XFR is key approach to perform zone data replication and synchronization,
The application of DRBD in Linux-HA Haibo Zhang 4/28/2014.
Comparison of Distributed Operating Systems. Systems Discussed ◦Plan 9 ◦AgentOS ◦Clouds ◦E1 ◦MOSIX.
Introduction to DFS. Distributed File Systems A file system whose clients, servers and storage devices are dispersed among the machines of a distributed.
7. Replication & HA Objectives –Understand Replication and HA Contents –Standby server –Failover clustering –Virtual server –Cluster –Replication Practicals.
Multiple Processor Systems. Multiprocessor Systems Continuous need for faster computers –shared memory model ( access nsec) –message passing multiprocessor.
1 Public DAFS Storage for High Performance Computing using MPI-I/O: Design and Experience Arkady Kanevsky & Peter Corbett Network Appliance Vijay Velusamy.
Presented By: Samreen Tahir Coda is a network file system and a descendent of the Andrew File System 2. It was designed to be: Highly Highly secure Available.
Chapter 11: File System Implementation Silberschatz, Galvin and Gagne ©2005 Operating System Concepts Chapter 11: File System Implementation Chapter.
COLO: COarse-grain LOck-stepping Virtual Machine for Non-stop Service Li Zhijian Fujitsu Limited.
The Mach System Silberschatz et al Presented By Anjana Venkat.
Chapter 7: Consistency & Replication IV - REPLICATION MANAGEMENT By Jyothsna Natarajan Instructor: Prof. Yanqing Zhang Course: Advanced Operating Systems.
Mellanox Connectivity Solutions for Scalable HPC Highest Performing, Most Efficient End-to-End Connectivity for Servers and Storage September 2010 Brandon.
Tweak Performance and Improve Availability of your Microsoft Azure VMs Rick
Distributed Computing Systems CSCI 6900/4900. Review Definition & characteristics of distributed systems Distributed system organization Design goals.
Considerations for Future Storage System Metrics and Efficiency Evaluation Green TWG - Design Metrics Subgroup 11/18/2015.
Background Computer System Architectures Computer System Software.
Distributed Systems: Distributed File Systems Ghada Ahmed, PhD. Assistant Prof., Computer Science Dept. Web:
Rick Claus Architect like a PRO for Performance and Availability of your Microsoft Azure VMs ARC43 6.
2015 Storage Developer Conference. © Intel Corporation. All Rights Reserved. RDMA with PMEM Software mechanisms for enabling access to remote persistent.
An Introduction to GPFS
Experiences with VI Communication for Database Storage Yuanyuan Zhou, Angelos Bilas, Suresh Jagannathan, Cezary Dubnicki, Jammes F. Philbin, Kai Li.
Parallel Virtual File System (PVFS) a.k.a. OrangeFS
Scaling Spark on HPC Systems
Alternative system models
Using non-volatile memory (NVDIMM-N) as block storage in Windows Server 2016 Tobias Klima Program Manager.
Persistent Memory over Fabrics
Microsoft Build /12/2018 5:05 AM Using non-volatile memory (NVDIMM-N) as byte-addressable storage in Windows Server 2016 Tobias Klima Program Manager.
Chet Douglas – DCG NVMS SW Architecture
Gregory Kesden, CSE-291 (Cloud Computing) Fall 2016
Chapter 1: Introduction
Real IBM C exam questions and answers
Chapter 12: File System Implementation
Chapter 7: Consistency & Replication IV - REPLICATION MANAGEMENT -Sumanth Kandagatla Instructor: Prof. Yanqing Zhang Advanced Operating Systems (CSC 8320)
Outline Midterm results summary Distributed file systems – continued
Multiple Processor Systems
Multiple Processor Systems
Multiple Processor and Distributed Systems
Basic organizations and memories in distributed computer systems
Presentation transcript:

Local and Remote byte-addressable NVDIMM High-level Use Cases Chet Douglas – Principal NVDIMM SW Architect - DCG 09-11-15

USE CASE – Local NVDIMM – Kernel Access -Memory Side Cache -Hierarchical Tiered Memory Client Workstation or Enterprise Server On Package High-Bandwidth Memory Hierarchical Tiered Memory and Caching -On-Package High-Bandwidth Memory caching DDR4 main memory -DDRT4 Main Memory caching byte-addressable NVDIMM -NVDIMM caching SSDs -Typical OS implementations will require extensive kernel access DDR4 Main Memory Byte addressable, direct access, NVDIMM SSD Asynchronous Demand On Package High-Bandwidth Memory Cache Requests Asynchronous Demand On Package High-Bandwidth Memory Cache Completions Asynchronous Demand DDR4 Main Memory Cache Requests Asynchronous Demand DDR4 Main Memory Cache Completions Asynchronous Demand Byte-Addressable NVDIMM Cache Requests (Writes, Flushes, Commits) Asynchronous Demand Byte-Addressable NVDIMM Cache Completions Asynchronous Demand SSD Requests Asynchronous Demand SSD Completions

USE CASE – Local NVDIMM – Kernel Access -SW RAID – Close “R5 Write Hole” -SW RAID – Cache for SSD RAID Volumes SSD RAID Volume Acceleration Cache Byte addressable, direct access, NVDIMM SSD RAID-5 Volume “Write Hole” Temp Data Store SW RAID Use Cases – Kernel Access -Typically SW RAID products utilize OS Kernel RAID Storage Drivers and UEFI RAID Pre-Boot drivers to implement SW RAID functionality. -Performance acceleration caches or caches to close vulnerabilities in RAID Data replication, require shared persistent kernel access to regions of the NVDIMM by both UEFI RAID Pre-Boot drivers and OS Kernel RAID Storage Drivers Byte addressable, direct access, NVDIMM Client Workstation or Enterprise Server SSD SSD RAID Volume Asynchronous Demand SW RAID Acceleration Cache Requests Asynchronous Demand SW RAID Acceleration Cache Completions Asynchronous Demand SW RAID Data Vulnerability Cache Requests Asynchronous Demand SW RAID Data Vulnerability Cache Completions Asynchronous Demand SW RAID Requests Asynchronous Demand SW RAID Completions

PUBLIC & PRIVATE CLOUD Client USE CASE – Local & Remote NVDIMM Public & Private Cloud -Synchronous (1:1) (active:passive) -Database Replication PUBLIC & PRIVATE CLOUD Active Server Node Active Server Node Database copies kept in lock step Client Active Database Master Synchronous Passive Database Slave Byte addressable, direct access, NVDIMM Byte addressable, direct access, NVDIMM Asynchronous Demand Database Master Updates Asynchronous Database Replication - RDMA Write Synchronous Database Replication SNIA Optimized Flush Synchronous Database Replication SNIA Optimized Flush Completion Asynchronous Demand Database Reads

USE CASE – Local & Remote NVDIMM Public & Private Cloud -Asynchronous (1:1) (active:passive) -Database Replication PUBLIC & PRIVATE CLOUD Active Server Node Database copies updated asynchronously as logfile updates are pushed to slave nodes Database Master Logfile Client Database Slave Logfile Active Server Node Active Database Master Asynchronous Passive Database Slave Byte addressable, direct access, NVDIMM Byte addressable, direct access, NVDIMM Asynchronous Demand Database Master Updates Local Logfile Update due to Database update Asynchronous Database Logfile Replication - RDMA Write Synchronous Database Logfile Replication SNIA Optimized Flush Synchronous Database Logfile Replication SNIA Optimized Flush Completion Local Database Update due to Logfile update Asynchronous Demand Database Reads

USE CASE – Remote NVDIMM Public & Private Cloud -Converged Disaggregated Network Topology -SW Defined Storage (SDS) PUBLIC & PRIVATE CLOUD Storage Processing Node Direct Attached NVDIMM Compute Node Virtual Clients (IaaS/PaaS) Byte addressable, direct access, NVDIMM Compute Node Asynchronous Demand SDS Updates Converged Disaggregated Network Topology  Compute Nodes and Storage Processing Nodes are physically separate nodes Asynchronous Storage Updates - RDMA Write Synchronous Storage Updates SNIA Optimized Flush Synchronous Storage Updates SNIA Optimized Flush Completion Asynchronous Demand SDS Reads

USE CASE – Remote NVDIMM Public & Private Cloud -Hyper-Converged Network Topology -SW Defined Storage (SDS) PUBLIC & PRIVATE CLOUD Compute Node Virtual Network Compute Node Virtual Clients (IaaS/PaaS) Virtual Storage Processing Node Compute Node Direct Attached NVDIMM Byte addressable, direct access, NVDIMM Asynchronous Demand SDS Updates Asynchronous Storage Updates - RDMA Write Hyper-Converged Network Topology  Compute Node and Storage Processing Node are in the same physically node Synchronous Storage Updates SNIA Optimized Flush Synchronous Storage Updates SNIA Optimized Flush Completion Asynchronous Demand SDS Reads

USE CASE – Remote NVDIMM Enterprise Appliance Enterprise Server Storage Processing Node Network Attached Enterprise Appliance Direct Attached NVDIMM Storage Attached Enterprise Appliance w Internal Network Fabric Switch & Direct Attached NVDIMM Enterprise Server Compute Node Byte addressable, direct access, NVDIMM Byte addressable, direct access, NVDIMM Asynchronous Storage Updates Asynchronous Appliance Updates - RDMA Write Synchronous Appliance Update SNIA Optimized Flush Synchronous Appliance Update SNIA Optimized Flush Completion

USE CASE – Remote NVDIMM HPC – PGAS, SHMEM, MPI, Human Brain Project -Distributed (1:N) (active:active) Database IO typically <= 256Bytes, Key/Value pair or pointer updates Active Server Node Synchronous Distributed Database Active Server Node Byte addressable, direct access, NVDIMM Distributed Database Reference Synchronous Distributed Database Byte addressable, direct access, NVDIMM Active Server Node Client Synchronous Distributed Database Byte addressable, direct access, NVDIMM Active Server Node Asynchronous Demand Database Updates Database is distributed across many nodes all actively taking part in Database references and updates Asynchronous Database Updates - RDMA Write Synchronous Distributed Database Asynchronous Database Replication SNIA Optimized Flush Asynchronous Database Replication SNIA Optimized Flush Completion Byte addressable, direct access, NVDIMM Asynchronous Demand Parallel Database Reads

USE CASE – Remote NVDIMM HA & HPC -Synchronous (1:N) (active:passive) -Database Replication w Parallel Access High Availability – Database copies kept in lock step Active Server Node Active Server Slave Node Synchronous Database Slave Client Active Server Slave Node Byte addressable, direct access, NVDIMM Database Master Synchronous Database Slave High Performance Computing - Parallel Database Reads typically load-balanced across all Passive Server Nodes Active Server Slave Node Byte addressable, direct access, NVDIMM Byte addressable, direct access, NVDIMM Synchronous Database Slave Asynchronous Demand Database Master Updates Byte addressable, direct access, NVDIMM Asynchronous Database Updates - RDMA Write Synchronous Database Replication SNIA Optimized Flush Synchronous Database Replication SNIA Optimized Flush Completion Asynchronous Demand Parallel Database Reads - RDMA Read