Data Center Conference 2014
Covers system design for data-center servers and networking
Held On February 5 - 6, 2014
Agenda for Day One: February 5, 2014
View Day Two
In this keynote, The Linley Group analysts provide insights into the important trends driving silicon for servers and networking, including microservers, silicon photonics, processor architectures, and macro forces. We examine how these trends impact the chips sitting at the heart of data-center equipment. We also provide a high-level view of the chip-vendor landscape, which remains surprisingly dynamic.
|10:00am-10:20am||BREAK – Sponsored by Freescale|
|10:20am-11:50am||Session 1: Server Technology
This session, moderated by Jag Bolaria, Principal Analyst at The Linley Group, will explore processors and technologies for data-center servers.
A Server Processor Optimized for Scale-Out Workloads
Scale-out workloads lend themselves to increasing aggregate application performance by employing multiple application instances working in parallel. This presents the opportunity to increase the performance of targeted applications, within a set of power and form-factor constraints, by utilizing more performance-power optimized processor cores instead of utilizing fewer, less power efficient cores with higher single-thread performance. This presentation will examine various factors involved in maximizing the performance of scale-out workloads within the power and form-factor constraints of a given chassis or rack.
Power Architecture Solutions for Scale-out Communications-Centric Data Center Processing
Communications systems look to take advantage of data center technologies such as virtualization, aggregation and Linux user space data path processing to solve demanding networking and security processing problems. Optimal solutions leverage multicore processing, integrated acceleration and networking interfaces. Freescale QorIQ processors provide leading performance, performance density, and cost and power savings while providing an Open Linux based user-space programming environment for networking and security application development. This presentation will examine this trend, discuss the QorIQ data path acceleration architecture and outline its advantages versus competitive solutions.
Developing Scalable and Resilient Memory Systems with Hybrid Memory Cube
Compute requirements in the data center are driving the need for higher memory bandwidth. Multiple memory technologies are being explored to address these needs. The Hybrid Memory Cube utilizes elements of 3-Dimensional Interconnect technology (through-silicon VIAs) and abstracted memory management as an efficient alternative to standard parallel memories. This discussion will compare and contrast the relative merits of DDR3, DDR4, and HMC with a focus on system complexity, reliability, and power consumption.There will be Q&A and a panel discussion featuring above speakers.
|11:50am-1:00pm||LUNCH – Sponsored by AppliedMicro Circuits|
|1:00pm-2:50pm||Session 2: Accelerating the Data Center
This session, moderated by Bob Wheeler, Principal Analyst at The Linley Group, will discuss how hardware accelerators can increase data-center performance and efficiency by offloading server CPUs.
Accelerating the Virtualized Data Center with Open vSwitch on TILE-IQ
Virtual switch implementations like OpenVswitch (OVS) provide VM-VM switching, VM isolation and VM migration capabilities along with OpenFlow protocol support. However, these capabilities come at a price: vSwitch on Xeon servers running multiple VMs consumes a significant fraction of CPU cycles. Unlike fixed-function NICs, Tilera's TILEncore-Gx Application Adaptors are fully-programmable PCIe cards using a standard Linux/C toolchain, and can offload complete applications from host servers. We'll describe the implementation and performance results of OVS running on a TILEncore Application Adaptor.
Networks in Transformation: SDN and NFV Implications on Processor Requirements
Hyper-scale data centers have led a transformation from, proprietary systems to software-controlled, commodity platforms for services and switching. New intelligent control points emerge in thousands of virtualized servers that are required to support multiple tenants, network overlays, and VM-demultiplexing, within a new security paradigm. These workloads consume compute resources and limit virtualized servers from scaling. A new class of processor is required to offload the embedded virtual switch to increase network performance and return valuable compute resources to applications.
Improving Cloud-Computing Efficiency using LiquidIO Application Accelerators
Cloud data centers are built around new virtualization technologies that improve operational efficiency. Network virtualization, however, also introduces significant overhead on server CPUs. Without effective offloads, the billable workload capacity of each server is reduced. In this talk, Cavium will show how its LiquidIO application accelerators can solve this challenge, freeing the server CPU to generate maximum services revenue while also minimizing opex.
FPGA-Based Acceleration in the Next-Generation Data Center
The tremendous compute horsepower, low-latency, and extreme power efficiencies of the FPGA have reshaped their role in the transformation of the next generation data center. FPGA based acceleration engines supported by C-based programming languages deliver orders of magnitude performance improvement over traditional processor based solutions for applications that include: next generation text and image search, financial trading, memcached, storage and encryption.There will be Q&A and a panel discussion featuring above speakers.
|2:50pm-3:10pm||BREAK - Sponsored by Freescale|
|3:10pm-4:40pm||Session 3: Advancements in Packet Processing
This session, moderated by Jag Bolaria, Principal Analyst at The Linley Group, will focus on the challenges of high-performance packet processing in the virtualized and SDN-enabled data center.
How VXLAN-Based Network Overlays Impact Ethernet IP in Data Center SoCs
As virtualized data centers continue to expand and evolve, Network Function Virtualization (NFV) is driving requirements for new network overlay protocols, such as Virtual Extensible LAN (VXLAN), running over Ethernet. This presentation reviews the benefits of VXLAN and how it can be used to overcome IP subnet limits. It will also cover the impact of VXLAN to Ethernet IP providers who enable designers to successfully implement the IP into their Data Center SoCs.
A New Programmable Paradigm for 40/100GbE Packet Processing
The disaggregation of computing, storage and networking in modern data center is resulting in ever-increasing bandwidth requirements as well as a need for standardization. These forces are accelerating the adoption of 40/100 GbE as the interconnect of choice in data centers faster than the merchant-silicon industry can react. Tabula's ABAX2 devices, developed on the leading-edge Intel 22-nm process, and rich suite of reference designs targeting line-rate packet processing at 40/100 GbE are uniquely positioned to solve this time-to-market dilemma.
Intelligent Offload Supporting Data Center Equipment at 100G+
MoSys will reveal the Third Generation Architecture of the Bandwidth Engine IC which delivers the high performance necessary for advanced Datacenter Networks. Firewall appliances have grown from single processor boxes to clusters of compute, route and inspection nodes which perform sophisticated security detection and prevention measures. Inside the datacenter the distributed nature of cloud content drives a many-fold times higher East-West bandwidth. Both applications are critically important to delivering a robust, low latency and high bandwidth user experience.There will be Q&A and a panel discussion featuring above speakers.
|4:40pm-6:10pm||RECEPTION AND EXHIBITS - Sponsored by AppliedMicro Circuits|