OSS Presentation Keynote by Hal Stern
- 1. The New Network is Programmable Hal Stern Chief Architect, Developer Business Unit Juniper Networks
- 2. Legal Statement This presentation sets forth Juniper Networks’ current intention and is subject to change at any time without notice. No purchases are contingent upon Juniper Networks delivering any feature or functionality depicted. The information and analysis provided in any Juniper Business Analysis are prepared for illustrative purposes only and should not be relied upon by participants as representative of actual results. Actual results of implementation of any Juniper Networks product or solution may vary based upon factors beyond the scope of analysis and/or beyond the control of Juniper Networks. Juniper Business Analysis does not constitute an offer, warranty or representation by or on behalf of Juniper Networks, Inc. and/or its subsidiaries, partners or affiliates
- 3. The Un-informed State Today THE HUMAN WORLD THE NETWORK WORLD Applications blindly probe the Networks spy on traffic to try network to understand what it to understand applications can deliver Network Aware Applications: Application Aware Networking: Game ping-stats, Doppler, Deep Packet Inspection, Geo-location, whois, Deep Flow Inspection proprietary codecs, Approximate application by proprietary control channels fingerprinting Approximate topology/location Service specific overlay topologies
- 4. A More Ideal State Bringing together the important players… … to enable interaction and influence the experience of the end user Application: Content: End-User: Network: Knows what the end-user Adjusts placement and Knows what it wants and Real-time interaction capabilities are. Informs selection of content to is directed there between application, placement of content in end-user, application content and end-users. proximity of end-user. desires and analytics. Enables multiple Control of resources. policy/profile points
- 5. What Is Possible? NETWORK APPLICATION APPLICATION NETWORK Applications made better by Networks made better by information information from network from application Understanding of end-device capabilities Bandwidth and resource optimization Real location / topology information New service topologies Adjust behavior to real-time usage or Protect network assets in real-time availability Closed user security groups Regulatory compliance Security identification Analytics of in-home usage Disambiguate sessions Billing granularity Disambiguate priorities Control of Resources from Flexibility of Service Placement Applications
- 6. Why Do You Care? • Developers love to tinker. --Cory Doctorow • Every single bit of storage is accessed over a network. • It’s time to start tinkering up and down the stack.
- 7. Tinkering With Storage • Drivers – SCSI, FC…. – Controller/general purpose system convergence • Filesystems – Performance – Naming – Redundancy
- 9. Storage In The Large == Networking In The Larger • Cost/Transit models for public/hybrid cloud • Security for data in motion • Congestion, latency and jitter are intentional • What does deep packet inspection show? – Popular storage buckets – Application access signatures – Latency sensitivity
- 10. Return of the Smart Network Voice Mostly Internet Converged Core 1975-1995 1995-2010 2010- Network is… Smart Dumb Smarter Clients are…. Really dumb Smart Smarter Data Rate 2.4 kb/s 10 Mb/s 100 Gb/s Defining immediate immediate immediate Factor connection access experience
- 11. Time and Functional Domains Analysis Modeling Action security Design posture QoS service level Real-Time enforcement cloud placement threat, Historical failure cloud cost analysis
- 12. It’s déjà vu all over again • Evolving APIs on the network software assets…. • Create new consumption and constraint patterns… • Requiring network-level information
- 13. Impact on Storage: Deutsch’s 8 Fallacies 1. The network is reliable 2. Latency is zero 3. Bandwidth is infinite Virtualization and cloud 4. The network is secure computing make bad assumptions worse 5. Topology doesn't change 6. There is one administrator 7. Transport cost is zero 8. The network is homogeneous
- 14. Information Creation • Point in Time • Annotation • Topology • Real Time • Context • Service Definition • Long Time • Corroboration • Metadata • Correlation • Analytics • Optimization
- 15. Decoupling of Control Points Policy Flows/Filtering Content/Resource Placement Delivery/Experience logs Applications statistics Networks ephemeral adapt to history adapt to networks state applications topology context
- 16. Closed-Loop System Action Value Configure Filter Provision GENERATE Correlate Dynamic Policy AFFECTS Combine INSIGHT Assets Data Inventory CREATE History Real-time Networking Devices, Servers, Databases, Applications etc.
- 17. Under The Hood NBI Bridge Java (and other languages) SOA, MTOSI via REST APIs & Libraries Interface Services users, RBAC, application deployment, UI, RESTful wrappers Element and Network Services Data Services inventory, topology, fabric, jobs, tagging, taxonomy, join/product, indexing, device management search/query Local Persistence DMI Database cluster
- 18. Making Networks Programmable 3rd party 3rd party applications & AppsSDKs & SDKs Enterprise SI Partner ISV Service Research Partner Provider Institution Customer Pulse - Endpoint Web Services APIs Space - Across the network CDN Data Center Control Management Data / Network APIs API API Packet API … … Junos SDK Network device
- 19. Content Request Routing Content Availability Per network response DNS Partner Content Load times and site availability Content Capacity New York Chicago Data Center Network Proximity Where do I get Answer: Network Availability DNS my content? Chicago! Network ISP A Congestion Media Flow Controller London London Data Center DNS ISP B BGP BGP Media Flow Controller Paris Network Data Server Amsterdam Data Center DNS ISP C Media Flow Controller
- 20. Bandwidth Calendaring Schedule a reserved path for your session… … without having to know the network. Real-time topology understanding Steering traffic through optimal paths Reservation transaction (WebServices API) Selecting specific traffic Web Services API Orchestration What would I use this for? Flexibility of service placement Network Scheduled data center backups Managed content distribution • Pre-positioning, live events Map-reduce class compute jobs • HPC data processing, search Cloud orchestration • Mobility of storage/VMs Location 1 Location 2 Scheduled and created application/session specific path through the network
- 21. JouleX Energy Manage (JEM) JouleX JEM Application 1 1 /device-management/devices 1 2 /device-management/devices/{device id} Junos Space Platform 1 3 /device-management/devices/{device id}/exec- rpc 1 1 Retrieve all devices under management API Native Apps 1 2 Get info for a single device managed by platform 1 3 RPC call to single device to get the power usage information. RPC for EX82xx: <get-power-budget-information> RPC for other Junos devices Platform <get-environment-pem-information> DMI/NetConf
- 22. Informing Applications Abstracted Network view • Map interface (network and cost maps) Network & Service data • Endpoint Cost query interface Geo location Subscriber Info Load/Utiliz Data Server ation Topology Clients TE Data (Network-aware Application-friendly protocol Source-specific protocols applications) • RESTful Web Service (Routing protocols, DMI, Data • JSON message encoding SNMP, WS, …) Sources
- 23. Network Application Domains • Content / Service Routing – Locate best copy of content for the end user, using customer rules • Managed content distribution – Content prepositioning to caches – Live events • Map-Reduce class of applications as consumers and producers – High-end distributed computing informed by and informing placement • Cloud OS network operations – Move VMs / Apps / Storage between locations • Cloudburst – Flexible on-demand allocation of cloud & network capacity • Security – DDoS attack prevention
- 24. Informing Control Points • What activities create the most traffic? • Where is latency introduced? Congestion? • What impacts user experience the most? • Extending network vocabulary to general purpose applications • Content networking as a future: – More about what, who, when of access, not where – Resource oriented: things you link, copy, incorporate and extend
- 25. What About “Big Data” “If you’re good at distribution (*), then you’re generating data. Use it.” Steve O’Grady, analyst, Redmonk “Big Data + Open Source = Big Money” (*) distribution of bits == networking
- 26. Data-Driven Use Case “Big Data” Business Constraints Hadoop Domain Real Time Analytics Java Developer Domain Ephemeral State Updates Tag/Filter/Join Configuration Event/Log Data Flow Data
- 27. Application Ecosystems 3rd Party 3rd Party Experience governed through App App data insight & ephemeral state control: make network aware of External Developer APIs applications Flow Content Subscriber Existing Systems Analysis Placement Analytics Junos Space APIs Make applications aware of network: Junos Space resources, integration points, services exposed Control Management Data / … … API API Packet API Network APIs Junos SDK Network device
- 28. Hard Problems… • Triangulating security, privacy and trust • Precision • Complexity – Running into potentially intractible (NP-Hard) problems with network structures and cost minimization • Time
- 29. Reach Info & Resources hstern@juniper.net @freeholdhal Developer Information & Resources www.juniper.net/developer @junosdeveloper helpmebuildmyapp@juniper.net