Tooling to Containerize an OpenCL Platform and Container Native Load Balancing
Details
Schedule
6:30 - 7:00 - Social / Food & Drinks
7:00 - 7:45 - Tooling to Containerize an OpenCL Visual Analytics Platform by Lee Butterman
Graphistry is a GPU-accelerated client/cloud visual investigation platform. Unusual for a GPU company, we have a policy of not touching hardware; instead, we deploy to the cloud or existing on-premise GPU clusters. By switching to Nvidia-Docker, we were able to turn deployments from an all-day affair to take just two minutes. The result is that we can now quickly deploy to users in a variety of environments, and we improved how our developers work on a daily basis. This talk describes the tools and processes we use, including our recently-released open source GPU container infrastructure layers that augment the core Nvidia-Docker tools.
Over the last year of on-premise deployments by our customers and use within our own cloud, we grew processes and tooling to address several key needs:
-
Automatic installation of standard runtimes and drivers;
-
Automatic validation of the runtime environment, especially for boxes not under our full control;
-
End-to-end testing around GPU use;
-
A CPU-only developer environment to ease use with other Docker tooling.
This talk shares how and why we achieved the above, and in a way in line with emerging Docker best practices. Along the way, we will dig into issues like surrounding tooling considerations such as supporting OpenCL, coordinating via Jenkins, and plugging into CI/Dockerhub/EC2.
7:45 - 8:30 - Enterprise Class Container Native Load Balancing by Chiradeep Vittal
Cloud-native, microservices-based / containerized applications are highly distributed over the network. Distributed applications complicate resilience, performance, security and visibility. Application architects have always used network proxies such as load balancers to add resilience, improve performance, block attacks and gain visibility into the network.
Cluster managers such as Docker Swarm and Kubernetes provide their own light-weight load balancers, usually integrated with the host kernel, for cross-container traffic while relying on traditional load balancers for Internet (ingress) traffic. Citrix NetScaler CPX provides a containerized load balancing solution suitable for both ingress and cross-container traffic for container clusters, with the same high performance and security of big iron load balancers.
Over the past several months in working with customers, Citrix has built capabilities in CPX to address the following needs:
(1) Integrating NetScaler CPX in the developer’s toolkit
(2) Policy-based networking to enable applications to be deployed into production
(3) Telemetry and app health scoring
8:30 - 9:00 - Networking / Wrap up
We look forward to seeing you there!
About the Speakers
Lee is currently an Infrastructure Engineer at Graphistry, a GPU-accelerated large-scale visual analytics startup. In his spare time he runs NoDictionaries.com, a Latin-reading website he made while in grad school for Classics. He lives in San Francisco.
Chiradeep Vittal is a Distinguished Engineer at Citrix Systems. Prior to Citrix, he was a founding engineer at Cloud.com (http://cloud.com/) (progenitor of Apache CloudStack -an IAAS platform), often tasked with the thorny details of virtualized networking and storage. Chiradeep has a B.Tech in Computer Science from IIT, Bombay and a M.Sc from the University of Alberta. He has spoken / presented at several conferences, including CloudStack Collab, LISA, OSCON, ONS, SDN Summit and LinuxCon.
