Upload
the-linux-foundation
View
557
Download
1
Tags:
Embed Size (px)
Citation preview
Copyright Cavium 2015
A Look Ahead: ARM Servers for Next-Generation Data Center & Cloud Infrastructure
Copyright Cavium 2015
61%
39%
49%
51%
37%
63%
2017 2016 2015 2014 2013 2012 0
20
40
60
80
100
120
140
160
180
200
Installed Servers in Cloud in Millions
CLOUD DATA CENTER (30% CAGR)
TRADITIONAL DATA CENTER (6% CAGR)
World is Increasingly Moving To The Cloud
Copyright Cavium 2015
2000 2005
2.8 ZB
72 ZB/yr
2014
Complexity increases exponentially with scale
Rapidly Expanding Physical Infrastructure
Copyright Cavium 2015
Virtual Platform
• Cloud Compute/Storage
• Network Virtualization (NFV)
• SDN
Server Market Faces Disruption
Public Clouds & Enterprise Servers
• 20% inc in shipment
in 2014
• Cloud driving growth in server shipments
ODM Direct & High Density Server
• Direct ODM model –
OpenCompute
• Scale out models
Workload Optimized Servers
• Targeted Application
• Optimized for
performance
Copyright Cavium 2015
Multiple applications consolidated in MULTI-TENANT SERVER FARM
ONE APPLICATION used by 10M+ USERS
The Drive for Workload Optimization
Copyright Cavium 2015
TRADITIONAL One Size Fits All Model.
Legacy Architectures are
stuck in the past Focused only around
CPU & memory
Lack of integration of
Workload Optimization
WORKLOAD OPTIMIZED
STORAGE NETWORK SECURITY COMPUTE
OPTIMIZED for COMPUTING
48 CORES
NETWORKING
MEMORY MEMORY
Copyright Cavium 2015
• Single threaded or limited multi-threaded program(s) • Structure query Databases / Email / Compilers/Interpreters
• Workload performance primarily dependent on
CPU/memory performance
• 1000s of applications used by 1000s of users
• Virtualization used to improve server utilization
• Managed by traditional IT
• Evaluated by traditional benchmarks
Legacy Server Applications
Copyright Cavium 2015
• Highly Distributed – the “System(s) are the Computer” Shared-nothing architectures - Distributed Data & Distributed Compute Many node environments Highly parallel – add more threads, go faster Multiple OS instances – fault tolerance in SW
• Big Data Large and highly distributed data sets
• Nodes are often special purpose
Cloud Applications can benefit from a “New Class of Servers” New class of servers requires new class of benchmarks
Cloud Applications = New Paradigm Of Computing
Copyright Cavium 2015
Workload Example/Use Case
Graph Search Social media data analysis (e.g. GraphLab, Giraph)
Web Caching Memcached
Media Serving Video server – e.g. “DASH” servers
Web Serving LAMP + Java/Tomcat/Ruby…
Data Analytics Hadoop (Mahout, Nutch)
Distributed Search Elastic Search
Distributed Storage Ceph (Object/Block) and HDFS (File)
Data Serving NoSQL type databases (e.g. Cassandra, Hbase, …)
Example Cloud Workloads
Copyright Cavium 2015
0
20
40
60
80
100
120
140
160
datacaching
dataserving
mapreduce
mediastreaming
web frontend
websearch
specint tpc-c tpc-e
SpecINT2006
Scaleout workloads
Source data from : A Case for Specialized Processors for Scale-Out Workloads Michael Ferdman, Almutaz Adileh, Onur Kocberber, Stavros Volos, Mohammad Alisafaee, Djordje Jevdjic, Cansu Kaynak, Adrian Daniel Popescu, Anastasia Ailamaki, Babak Falsafi, In IEEE Micro's Top Picks, 2014
Cloud Workloads are Different Example #1 – Very different instruction miss rates
Mis
ses
Per 1
K In
stru
ctio
ns (M
PKI)
MPKI
Copyright Cavium 2015
00.20.40.60.8
11.21.41.6
Cloud Workloads are Different Example #2 – impact of IPC changing
Inst
ruct
ion
Per C
ycle
(ipc
)
CPU Intensive Traditional Benchmarks
Source data from : A Case for Specialized Processors for Scale-Out Workloads Michael Ferdman, Almutaz Adileh, Onur Kocberber, Stavros Volos, Mohammad Alisafaee, Djordje Jevdjic, Cansu Kaynak, Adrian Daniel Popescu, Anastasia Ailamaki, Babak Falsafi, In IEEE Micro's Top Picks, 2014
Copyright Cavium 2015
Cloud Workloads are Different Example #3 – Performance Sensitivity LLC & L2 Caches
Source data from : A Case for Specialized Processors for Scale-Out Workloads Michael Ferdman, Almutaz Adileh, Onur Kocberber, Stavros Volos, Mohammad Alisafaee, Djordje Jevdjic, Cansu Kaynak, Adrian Daniel Popescu, Anastasia Ailamaki, Babak Falsafi, In IEEE Micro's Top Picks, 2014
Copyright Cavium 2015
• Optimum Choice and size of Caches are different for scale out workloads • Lower IPC Less parallelism available Aggressive, out-of-order, wide issue machines don’t benefit
Don’t waste area and power
• Scaleout Highly parallel nature, more independent processing cores.
Use the area for more cores Large number of more efficient cores provide lower power and more
performance for Scale Out Workloads
Implications to Processor Design
Copyright Cavium 2015
Can fit multiple cores in area of one complex core
For Scale Out workloads, multiple cores provide the best performance/unit area / watt
Complex Single Core Multiple Cores
Drive to Efficiency
Copyright Cavium 2015
• ARM designs and licenses IP Standard License Architectural License
• ARM enables ecosystem via Foundation Models Reference Systems
• ARM designs and sells tools and supporting collateral
ARM Engineering Model
Copyright Cavium 2015
• Focused on CPU and memory (scalar performance)
• Very little network or disk IO
• Not a distributed application
• No cooperative tasks, no sharing, no communication
• No operating system or Hypervisor impact
• Many of the sub elements are cache friendly
• Could be very expensive to setup and run
• Some need very big systems
Characteristics of Traditional Benchmarks
Traditional Benchmarks do not adequately represent Cloud Workloads
Copyright Cavium 2015
Workload Focused Support large & distributed Data Sets Scalable across many nodes & cores
Different Behavior from Transaction & CPU Memory SoC includes more “System” features
Need to expand beyond CPU & memory and include network & storage IO
New Scale Out Benchmarks
Copyright Cavium 2015
open source
open source
open source
open source
open source
open source
open source
open source
open source
open source
open source
open source
open source
open source
open source
open source
open source
open source
open source
open source
open source
open source
open source
open source
open source
open source
open source
open source
open source
open source
open source
open source
open source
open source
open source
open source
open source
open source
open source
open source
open source
open source
open source
open source
open source
open source
open source
open source
open source
OPEN SOURCE
OPEN LICENSES
OPEN INNOVATION
OPEN DISTRIBUTION
OPEN COMPETITION OPEN PARTNERSHIP
OPEN DATA STANDARDS OPEN OPPORTUNITIES
OPEN TECHNOLOGIES
Open Source is Expanding
Copyright Cavium 2015
Standards – Core to Innovation and Scale
The broader industry committed and engaged
Copyright Cavium 2015
● Formed in November 2012 ● Working on core open-source
software for ARM servers ● Boot architecture – UEFI/ACPI ● Virtualization – KVM/Xen ● ARMv8 bringup & optimization ● LAMP, OpenJDK, Hadoop, OpenStack
● Reduces costs, eliminates fragmentation, accelerates time to market
● Members can focus on innovation and differentiated value-add
Linaro Enterprise Group (LEG)
http://wiki.linaro.org/LEG
Copyright Cavium 2015
Summary Infrastructure HyperGrowth Server Disruption
Workload Optimization Rapidly Growing EcoSystem