Implementing a unified Datacenter architecture for service providers IT, commercial & network Tomas Fredberg, Ericsson HDS 8000 Chief Architect
Content Introduction Ericsson HDS 8000 Hardware components Workload optimization Interfaces Datacenter evolution Command Center interfaces in relation to Rackscale Architecture Reflections Summary Intel Developer Forum August 2015 Page 2
Ericsson hds 8000 supports all workloads Telecom Cloud IT Cloud Commercial Cloud Network functions OSS/BSS, Media & IT functions Commercial XaaS offerings Cloud Infrastructure Governance & Security Cloud Infrastructure Automation Services Service-Provider Software Defined Networking BSP HDS Intel Developer Forum August 2015 Page 3
Multi-application platform history of HDS 8000 Started development of multi-application infrastructure systems 1st generation based on Blade System (Ericsson Integrated Site) 2nd generation platform (Ericsson BSP 8100) 3rd generation cloud and datacenter optimized system (Ericsson HDS 8000) 2002 2005 2010 Now Intel Developer Forum August 2015 Page 4
Ericsson HDS 8000 - Hyperscale Computing For Everybody Next generation datacenter infrastructure platform Built in collaboration with Intel Disaggregated datacenter solution using Intel Rack Scale Architecture Fully optical interconnect Intel Developer Forum August 2015 Page 5
Ericsson HDS 8000 CPU RAM NIC Accelerators Switch Disk Disaggregated hardware Seamless scalability with efficient life cycle management Intel Developer Forum August 2015 Page 6 Full optical interconnect Enabling hyperscale Futureproof deployments HDS Command Center Advanced analytics, automation, orchestration and asset governance
Flexibility Today s Forced Choice: flexibility vs. utilization high HDS Standard/Generic Solutions Software Defined Flexible Solution Enables High Utilization Not Optimized Solutions Specific/Dedicated Solutions low low Utilization high Intel Developer Forum August 2015 Page 7
HDS 8000 components Abstract Software Defined Infrastructure Physical Resources Composed vpod 1 Hardware Compute Pool Disk Pool vpod M Command Center software Framework: Physical Enclosure, Management Channels, Fiber Backplanes Switch Pool Acceleration Pool Intel Developer Forum August 2015 Page 8
HDS Workload optimization Software defined vpods created from heterogeneous hardware pools Workload 1 placed Workload in N Domain placed 1in Domain 1 Workload 1 placed Workload in K Domain placed Min Domain M Native Workload Domain 1 Cloud Manager VMM 1 VMM M Network Management System vpod 1 vpod M vpod X Ericsson HDS 8000 HDS components HW pools HW systems Intel Developer Forum August 2015 Page 9
HDS Interfaces Supports all popular requested OS environments and APIs (Illustration shows some examples of supported interfaces) Workload 1 placed Workload in N Domain placed 1in Domain 1 Workload 1 placed Workload in K Domain placed Min Domain M Native Workload Domain 1 Cloud Manager VMWare VMM 1 Openstack VMM M Network Management System VMWare API vpod 1 OpenStack API vpod M Open REST API vpod X Ericsson HDS Command Center Open REST API + GUI Agents PSME DMTF, SNMP, Console, HW systems etc. Intel Developer Forum August 2015 Page 10
One chassis for many generations of server and storage sleds Enabled by a future proof passive optical backplane and self contained sleds with a simple interface: Protocol agnostic opto links for all user data Standard AC feed Standard equipment management interface Intel Developer Forum August 2015 Page 11
External Optical interface Connector field for multi ribbon fibers Fan-Out fiber cable Intel Developer Forum August 2015 Page 12
HDS Optical Interfaces Flexibility Pooled disk system example SAS SATA NVMe SAS SATA NVMe NVMe+ SAS NVMe+ SATA SAS Expander Optics PCIe SAS g3 switch/expander Expander Optics PCIe g4 SAS switch/expander Expander Optics+ Cabled Optical Backplane Optics SAS Controller Optics PCIeSAS g3 switch/controller Controller Optics+ SAS Controller PCIe g4 switch/controller Intel Developer Forum August 2015 Page 13
HDS datacenter Evolution New Features & Components (e.g. CPU, Disk, RAM, Acceleration, Switch) Orchestrated Hardware Components Feedback for optimization Application Analytics vpod vpod Component Pools Software Defined Infrastructure Infrastructure Analytics Software-defined infrastructure enables optimization based on real time analytics Intel Developer Forum August 2015 Page 14
HDS Command Center interfaces RSA compliant APIs with value add extensions RSA NB Standard API Ericsson NB Extension API HDS Command Center POD Management Intel RSA POD manager used as services Ericsson HDS 8000 value add? Ecosystem added value PSME Adaptor (Uses parts of Intel PSME functions) PSME IPMI/SNMP/Console Proprietary Proprietary 3PP Box e.g. Server PSME Pooled System 3PP Box e.g. Server Intel Developer Forum August 2015 Page 15
SDI reflections Industry wants software defined infrastructure Intel Rack Scale Architecture / HDS delivers! Automation and higher level services needed to provide good value Aim to simplify for users - but that is not simple For the Ecosystem and SDI users: How can we offer high level services that also simplify usage for the users? Which interfaces and models make service delivery simpler and more efficient? Where should these interfaces and models be specified and standardized? Intel Developer Forum August 2015 Page 16
Utilization reflections Industry wants higher utilizations - Intel Rack Scale Architecture / HDS delivers through shared resources / pooled systems The key requirements to enable success A thorough security architecture Simple handling and interaction of Multi pool systems e.g. user administration, certificates, fault indications Multi user systems e.g. Service Level Agreements, resource allocation Good analytics Higher abstraction levels of user API and models Standardization of metrics for more comparable systems Intel Developer Forum August 2015 Page 17
Summary Ericsson HDS 8000: Rackscale compliant Software Defined Infrastructure with 3PP support vpod extensions offer software defined POD hardware partitions with analytics Fiber optics for scalability and future proof installations Ecosystem contribution to Intel Rack Scale Architecture success should aim for automation and simplicity using modular and portable designs Consumption of infrastructure on higher abstraction level gives higher efficiency but requires use pattern changes Intel Developer Forum August 2015 Page 18