Cavium joined Red Hat Storage Day New York on 1/19/16 to give the history of the ARM server ecosystem, explain the innovation of ThunderX, and describe scale out's influence on target workloads.
6. Copyright 2014 Cavium. Confidential.
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
By now you have seen that we have set the bar very high for ThunderX in both performance and specific workload optimization. Now in order to make ThunderX the ultimate solution for the hyperscale data center and cloud we are also setting the bar very high for system enablement, our software development ,our leading partner engagements and the overall industry ecosystem. Today I will cover all of these areas.
First I want to give you a perspective on the growth that has been underway in the ARM Server ecosystem. Over the past 6 years we have seen the momentum continuing to build around ARM Servers in key areas that are the core for a thriving ecosystem. These include
Platform Standards – such as the Server Base System Architecture
Industry Consortia – OpenStack and OpenCompute and Linaro
Open Source Community – obviously the Linux kernel and os but also tools and applications
Linux OS Distributions – adoption across the board
Development Tools – not just compilers and tools but interpretive languages such as Java
Scale Out Applications – Ceph storage
Today with Thunder we are happy to expand this significantly
Let us now talk about performance. Performance & performance per watt are both extremely important metrics for server customers.
We look at performance from two standpoint – traditional server benchmarks that are mainly focused around CPU and memory – examples include SPECInt_rate that measures integer performance, lmbench and stream that measure memory performance as well as performance for our target cloud applications.
On traditional server benchmarks, CN88xx delivers performance comparable to main stream server processors at significantly lower cpu subsystem power. CN87xx delivers performance comparable to micro server processors such as Avoton, AMD Seattle and APM X-Gene.
However for target cloud workloads such as web caching (e.g. Memcached), web serving, Data analytics (e.g. Hadoop), https server etc., ThunderX’s combination of high core count, integrated hardware accelerators coupled with full system level virtualization support and integrated IO deliver much better performance at significantly lower system power.
X86 Servers based on Single 2630v3 - ThunderX Servers based on Single 8890 (48-core).
RADOS Bench Read/Write – same block size, replication (3), servers, OSDs