Presenter: Hung-Fu Li HPDS Lab. NKUAS 2009-12-31 1 vCUDA: GPU Accelerated High Performance Computing...
-
Upload
nicholas-parsons -
Category
Documents
-
view
212 -
download
0
Transcript of Presenter: Hung-Fu Li HPDS Lab. NKUAS 2009-12-31 1 vCUDA: GPU Accelerated High Performance Computing...
Presenter: Hung-Fu Li
HPDS Lab.NKUAS
vCUDA: GPU Accelerated High Performance Computing in Virtual Machines
Lin Shi, Hao Chen and Jianhua Sun
IEEE 2009
2
Lecture Outline
Abstract 3Background 4Motivation 5CUDA Architecture 7vCUDA Architecture 8Experiment Result 13Conclusion 19
3
Abstract
This paper describe vCUDA, a GPGPU computation solution for virtual machine. The author announced that the API interception and redirection could provide transparent and high performance to the applications.This paper would carry out the performance evaluation on the overhead of their framework.
4
Background
VM(Virtual Machine)CUDA (Computation Unified Device Architecture)API (Application Programming Interface)API Interception, RedirectionRPC(Remote Procedure Call)
5
Motivation
Virtualization may be the simplest solution to heterogeneous computation environment.Hardware varied by vendors, it is not necessary for VM-developer to implements hardware drivers for them. (due to license, vendor would not public the source and kernel technique)
6
Motivation ( cont. )
Currently the virtualization does only support Accelerated Graphic API such as OpenGL, named VMGL, which is not used for general computation purpose.
7
CUDA Architecture
Component Stack
CUDA Enabled Device
CUDA Driver API
CUDA Runtime API
CUDA Driver
User Application
<< CUDA Extensions to C>>
8
vCUDA Architecture
Split the stack into hardware/software binding
CUDA Enabled Device
CUDA Driver API
CUDA Runtime API
CUDA Driver
User Application
<< CUDA Extensions to C>>
hard binding
soft binding
Direct communicate
Part of SDK
9
vCUDA Architecture ( cont. )
Re-group the stack into host and remote side.
CUDA Enabled Device
[v]CUDA Driver API
[v]CUDA Runtime API
CUDA Driver
User Application
<< CUDA Extensions to C>>
CUDA Driver API
Host binding
Remote binding(guestOS)
Part of SDK
[v]CUDA Enabled Device(vGPU)
10
vCUDA Architecture ( cont. )
Use fake API as adapter to adapt the instant driver and the virtual driver.API Interception
Parameters passed
Order Semantics
Hardware State
Communication
Use Lazy-RPC TransmissionUse XML-RPC as high-level communication.(for cross-platform requirement)
[v]CUDA Driver API
[v]CUDA Runtime APIRemote binding(guestOS)[v]CUDA Enabled Device(vGPU)
11
vCUDA Architecture ( cont. )
Virtual Machine OSHost OS
lazyRPC
Non instant API
Instant API
12
vCUDA Architecture ( cont. )
vCUDA API with virtual GPULazy RPC
Reduce the overhead of switching between host OS and guest OS.
AP LazyRPC
vGPUHardware states
API Invocation
GPU
Instant api call
NonInstant API call
NonInstant Package
Stub
vStub
13
Experiment Result
CriteriaPerformance
Lazy RPC and Concurrency
Suspend& Resume
Compatibility
14
Experiment Result ( cont. )Experiment Result ( cont. )
CriteriaPerformance
Lazy RPC and Concurrency
Suspend& Resume
Compatibility
15
Experiment Result ( cont. )Experiment Result ( cont. )
CriteriaPerformance
Lazy RPC and Concurrency
Suspend& Resume
Compatibility
16
Experiment Result ( cont. )Experiment Result ( cont. )
CriteriaPerformance
Lazy RPC and Concurrency
Suspend& Resume
Compatibility
17
Experiment Result ( cont. )Experiment Result ( cont. )
CriteriaPerformance
Lazy RPC and Concurrency
Suspend& Resume
Compatibility
18
Experiment Result ( cont. )Experiment Result ( cont. )
CriteriaPerformance
Lazy RPC and Concurrency
Suspend& Resume
Compatibility
MV: Matrix Vector Multiplication AlgorithmStoreGPU: Exploiting Graphics Processing Units to Accelerate Distributed Storage Systems MRRR: Multiple Relatively Robust RepresentationsGPUmg: Molecular Dynamics Simulation with GPU
19
Conclusion
They have developed CUDA interface for virtual machine, which is compatible to the native interface. The data transmission is a significant bottleneck, due to RPC XML-parsing. This presentation have briefly present the major architecture of the vCUDA and the idea of it. We could extend the architecture as component / solution to make the cloud computing support GPU.
20
End of Presentation
Thanks for your listening.