Partitioned Multistack Evironments for Exascale Systems

Jack LangeAssistant ProfessorUniversity of Pittsburgh

What I’ve heard…

• In-situ everything…– Network will not support current behaviors– Must collapse multiple functions onto single platform

• Visualization, data analysis, checkpointing, etc…

VisualizationCluster

Supercomputer

StorageCluster

Exascale Machine

What does this mean for the OS?• At Petascale we could optimize each environment separately

– Each had their own OS and hardware• At Exascale workloads will be co-located

– Can a single OS handle all workloads effectively?

• Claim: Probably not– Each has different resource requirements and behaviors– Exascale will need to support multiple OS environments on the same hardware

Exascale Node

Lightweight Kernel

HPC applicationManagement

Processes

Resource Manager

EmbeddedLinuxLinux

Analysis+

Visualization

Linux

Debugger

Challenges

• Increase in complexity at both hardware and software layers– Heterogeneous hardware

• GPUs, Lightweight cores, SSDs, …

– Complex Topologies• NUMA on chip, NUMA on node, dedicated GPU nodes, …

– Heterogeneous applications– Hardware and software failures– Power constraints

• How can we manage all of this at the OS layer?– A unified and monolithic OS environment isn’t going to work

Approaches

• Exascale machines will need the ability to run multiple OS instances in parallel– Each targeting a particular application/workload

• Linux for vis., LWK for application, etc

• Hopefully virtualization can help…– But there will probably be limited hardware support for it

• Will need other techniques for partitioning resources– Virtualization-lite?– Lightweight and distributed resource managers– Flexible communication channels– Many others…