Support for Managed-Language Workloads in Warehouse-Scale Computers
Event details
| Date | 01.12.2015 |
| Hour | 12:00 › 13:00 |
| Speaker | Martin Maas, University of California, Berkeley |
| Location | |
| Category | Conferences - Seminars |
Data center applications and frameworks such as Hadoop, Spark and ZooKeeper are frequently written in managed languages such as Java or Scala. However, the managed language runtime systems they are running across are not primarily designed for distributed data center workloads and therefore exhibit inefficiencies when applied in this setting.
At the same time, the design of warehouse-scale computers is changing. We are expecting a shift towards rack-scale units with custom SoCs, fast interconnects and large amounts of shared memory. We believe that this trend creates opportunities to improve support for managed-language workloads in data centers.
In this talk, I will present some of these challenges and opportunities. I will then show how we address some of the challenges in what we call a "Holistic Language Runtime System”, a distributed language runtime that collectively manages runtime services across multiple nodes. Using Garbage Collection (GC) as an example, I will show how a Holistic Runtime System is effective both in reducing the impact of GC pauses on a big data workload, and in improving GC-related tail-latencies for a latency-sensitive workload.
At the same time, the design of warehouse-scale computers is changing. We are expecting a shift towards rack-scale units with custom SoCs, fast interconnects and large amounts of shared memory. We believe that this trend creates opportunities to improve support for managed-language workloads in data centers.
In this talk, I will present some of these challenges and opportunities. I will then show how we address some of the challenges in what we call a "Holistic Language Runtime System”, a distributed language runtime that collectively manages runtime services across multiple nodes. Using Garbage Collection (GC) as an example, I will show how a Holistic Runtime System is effective both in reducing the impact of GC pauses on a big data workload, and in improving GC-related tail-latencies for a latency-sensitive workload.
Practical information
- Informed public
- Free
Organizer
- Babak Falsafi
Contact
- Stéphanie Baillargues