Emulex I/O Solutions Storage Trends 2014 Hilmar Beck Senior Sales Engineer Next Generation Data Center Trends Enterprise Virtualization Cloud Network Security Application Delivery Virtualization and consolidation growth continues unabated and drives I/O bandwidth Cloud strategies and converged networks require technology and tools to measure quality of service and guarantee it Increasing predatory attacks on IT portals press need for security solutions and forensics requires capture solutions Application specific performance demands are driving latencies and movement to 10/40/100Gb converged fabrics Flash Storage Deployment of SSDs in storage arrays is pushing networking bandwidth and latency IT Trends Driving New Networking Requirements 2 Emulex Confidential - © 2013 Emulex Corporation What is Gen5 Fibre Channel? Speed-based naming changed to generational-based naming New advanced protocol services and features running multiple speeds – Not just bandwidth improvement – 16GFC, 8GFC, 4GFC speeds – Maintains backward compatibility to previous FC generations 1Gb 2Gb 4Gb 8Gb Gen 5 Gen 6 Fibre Channel Fibre Channel Fibre Channel Fibre Channel Fibre Channel Fibre Channel Gen 1 Gen 2 Gen 3 Gen 4 16Gb 32Gb 3 Emulex Confidential - © 2013 Emulex Corporation Advanced features Multi-speed 16GFC/8GFC/4GFC Emulex ExpressLane™ SSD Latency Challenge • Flash storage shares the SAN with traditional rotating media • • 4 QoS Solution • ExpressLane creates separate queues for low latency storageIdentified by LUN Mission-critical requests are stuck behind requests to slow storage • Individual queues are coalesced for latency not overall bandwidth Current queuing mechanisms are optimized for throughput, not latency • Queue associations are made from OneCommand Manager Emulex Confidential - © 2013 Emulex Corporation Emulex ExpressLane™ 1. VM1 on Server 1 requests low priority I/O from Disk Array 1 VM1 VM2 Server 2 Server 1 Congestion Point 3 VM1 Traffic 1 Disk Array 5 VM2 Traffic 2. VM2 requests high priority IO from Disk flash array 3. Congestion occurs at HBA - VM2 traffic is stuck behind slower traffic 2 Disk/ Flash Array Congestion at the HBA (multiple VMs requesting I/0) Emulex Confidential - © 2013 Emulex Corporation ExpressLane prioritizes VM2 traffic over VM1 Provides Quality of Service, I/Os have consistent latency performance Emulex CrossLink™ SSD Isolation Challenge • Flash requires coordination between nodes • Current solutions (TCP/IP and UDP) suffer from “roundabout” stack-hopping • • • 6 SSD Coordination Solution • In-band FC Messaging solves latency & “stack-hopping” for cache or device coordination Latency & QoS issues over Ethernet hamper coordination with storage devices • Uses standard & proven FC-CT protocol for FC and FCoE Trust issues (storage networks deemed implicitly secure) • Simple interface- kernel or API Separate Ethernet connectivity requires additional wiring & management Emulex Confidential - © 2013 Emulex Corporation Emulex CrossLink™ Example: VM Migration Cache Prefill Server B (Destination VM) Server A (Source VM) 3 1. Cache Prefill 1 2. Server A also sends cache meta data via Crosslink 3. Caching software in Server B processes metadata to create a “to do” list of pre-fill operations 4. Caching software issues standard read operations to load cache 2 4 Tiered Array 7 VM migrates from Server A to Server B Cache Prefill (Virtual Machine Migration) Emulex Confidential - © 2013 Emulex Corporation Emulex & Brocade ClearLink (D_Port) Support ClearLink is a rich SAN diagnostic mode from Brocade Gen 5 FC switches (16GFC) ClearLink is now supported by all 16G (only) host-based Emulex LightPulse Gen 5 FC HBAs (XE201 based) Identifies and isolates marginal link level failures and performance issues: SFP, port, & cable Emulex # 1 Gen 5 HBAs + Brocade #1 Gen 5 switches, Together Providing Superior SAN-wide Diagnostics 8 Emulex Confidential - © 2013 Emulex Corporation ClearLink D_Port Saves Time & Money Troubleshooting A cable is faulty, but how do you find it? Can be hundreds of cables & SFPs in the SAN. You could start replacing cables and SFPs, one by one using trial & error Wastes time – Or replace all cables and SFPs in the environment and try again Expensive & wastes time 9 Emulex Confidential - © 2013 Emulex Corporation Emulex Gen 5 Fibre Channel Advanced Features for Best Flash/Cache, VMs Emulex ExpressLane™ Emulex CrossLink ™ ClearLink Enablement Priority Queuing Quality of Service (QoS) & performance to meet SLAs for application sensitive data, flash/cache In-band Message-passing Rich Diagnostic D_port Significantly reduces latency, improves CPU utilization Reduces downtime, saves time & money troubleshooting problems Alleviates congested networks Increases reliability Simplifies management Maximizes ROI on Flash/cache systems 10 Emulex Confidential - © 2013 Emulex Corporation Industry-leading reliability Emulex Gen5 Accelerates Application Performance Versus 8GFC 41% 33% Faster workload completion for SQLServer Database Applications Data Warehousing Workload Virtualization / Cloud Exchange Workload 75% 11 More Transactions per Second for SQLServer (100-400 users) Better throughput VMware ESXi 3x Emulex Confidential - © 2013 Emulex Corporation IOPS vs. 8GFC for Exchange Ethernet Connectvity Discrete Networking Converged Networking Software Defined Convergence – 3X Cost – FCoE Driving 10GbE – RDMA over CE (RoCE) – 3X Management – Virtual Networking – Application Acceleration – 3X Cables – Led by Blade Servers – Virtual I/O (OVN, SR-IOV) – 3X Switching – Telco and Web Giants – Driving 40 and 100GbE – Cloud, HPC, Big Data 12 Unique “switch agnostic” positioning Emulex Confidential - © 2013 Emulex Corporation Performance - What is New with the OCe14000? Cloud, Big Data & SDN VNeX Virtualization 70% Faster Hybrid Cloud 50% Better CPU Efficiency Secure Multi-Tenant Clouds SDN & Workload Optimization Hybrid Cloud NVGRE/VXLAN 4X Small Packet Performance 50% Better IOPS Lowest CPU Utilization Save up to 50W/Server Highest Bandwidth & IOPS / Watt Web-scale Performance 13 Operational Efficiency Emulex Confidential - © 2013 Emulex Corporation High Performance Networking Skyhawk RoCE What is Remote Direct Memory Management (RDMA)? What is RDMA over Converged Ethernet (RoCE)? Benefits of Skyhawk-R with RoCE Ability to remotely manage memory enabling server to server data movement directly between application memory without any CPU involvement A mechanism to provide this efficient data transfer with very low latencies on Ethernet Basically IB Over Ethernet (with PFC, etc.) Classic Ethernet is a Best Effort Protocol Delivers low latency for performance-critical and transaction intensive applications Better OPEX vs. Infiniband infrastructure which is requires a unique fabric which is difficult to deploy and manage RoCE is IB Over Existing Network Infrastructure 14 Emulex Confidential - © 2013 Emulex Corporation Emulex Confidential © 2013 Emulex Corporation 14 Enterprise Cloud Needs RDMA RDMA Read Efficiency Gains vs. TCP/IP 1200% 1000% 800% 600% 400% 200% 0% 512B 1K 2K 32K 64k 128k 256K Internal testing - Emulex OCe14000 using SMB 3.0 on Windows Server 2012 R2 15 Emulex Confidential - © 2013 Emulex Corporation 512K 1M Importance of File Transfer Performance RDMA delivers 77% faster transfers Our pockets are generating Big Data – Growing amount of digital content on mobile devices 35 30 25 20 15 10 5 0 Transfer Time for 12GB file RDMA TCP/IP Internal Testing: Emulex OCe14000 Using OFED 3.5.2 16 April 2-3, 2014 Emulex Confidential - © 2013 Emulex Corporation #2014IBUG Move to Software-defined Convergence Multi-Fabric Block I/O (FC, FCoE & iSCSI) 6% CAGR Through 2016* Software-Defined Convergence 25% 10/40G CAGR Through 2016* Sources: *Crehan Research: Server-class Adapter and LOM Controller Long-range Forecast, July 2013 and **Dell’Oro Group: Fibre Channel Adapter Vendor Report 2Q2013, Aug. 2013 17 Emulex Confidential - © 2013 Emulex Corporation Emulex RoCE Offerings XE100 Series 10/40GbE Network Controller OCe14101/2 10GbE SFP+ Ethernet Adapters OCe14401 40GbE QSFP+ Ethernet Adapters 18 Emulex Confidential - © 2013 Emulex Corporation Where to use RDMA Initial Applications: – Windows 2012 SMB/Direct • Which implies SQL/Server, Hyper-V VM migration etc. – Linux NFS/RDMA Microsoft will actively market Windows R2 and SMB/Direct – Better CPU efficiency, faster VM migrations, etc. – Convince the E/U they need or might want the option Others will be added in the future – Based on OEM, Partners and E/U requests and appropriate business case 19 Emulex Confidential - © 2013 Emulex Corporation RoCE in Action Latency Application Application User Buffer User Buffer TCP TCP UDP UDP Bypassed Latency IP IP Latency Standard 10GbE Adapter Outgoing Data Incoming Data RDMA enabled 10GbE Adapter Outgoing Data Without RoCE 20 20 Incoming Data With RoCE Emulex Confidential - © 2013 Emulex Corporation Thank You! 21 Emulex Confidential - © 2013 Emulex Corporation