High Performance Computing on Virtualized Environments Ganesh Thiagarajan Fall 2014 Instructor: Yuzhe(Richard) Tang Syracuse University

Overview  Virtualization refers to act of creating a virtual version of hardware or software  Virtual machine is an emulation of a computer system  In this paper, we are mainly discussing about  Virtualization,  how to achieve virtualization in Higher Performance Computing applications  XEN which is a Hypervisor  various other factors that affect achieving high performance in Virtualization

Introduction  Virtualization:  a technique which utilizes the underlying hardware for creating different many copies of virtual machines  Benefits  Reduction of power  Cooling costs  Increased availability  Isolation  Reduce Hardware Vendor lock-in  Every Virtual Machine is a distinct copy  Depending on the availability of Hardware Resources, a number of Virtual Machines can be allocated  Allocating virtual machines to available hardware is called Virtual Machine Placement Problem

Virtualization (Cont.)  Virtualization modes  Full-Virtualization  Para-Virtualization Full VirtualizationPara Virtualization

Virtualization (Cont.)  Full Virtualization:  Advantages:  improves CPU and performance  gives the flexibility of running different Operating Systems  Live migration is very easy  Disadvantages:  Hardware and Software components needs to be compatible  Due to x86 protected instruction, performance might be low

Virtualization (Cont.)  Para Virtualization:  Advantages:  very easier to implement  Considerable increase in the performance  Disadvantages:  Guest OS requires modification which in turn results in poor compatibility

Virtual Machine Placement  Bin Packing solution problem  Offline Algorithms  Online Algorithms  Next Fit Algorithm  First Fit Algorithm  Best Fit Algorithm

Virtual Machine Live Migration  can be migrated from one hardware to other available hardware resource  without disturbing the running virtual machine  without shutting down the Operating System  can be performed by Hypervisor  Sample Live Migration Code:

Virtual Machine Monitor  Software program which runs over the hardware is Hypervisor  Types:  Type 1 Hypervisor  Type 2 Hypervisor Type 1 HypervisorType 2 Hypervisor

Virtual Machine Monitor (Cont.)  a piece of software which manages Virtual Machines  also termed as Hypervisor  Examples: Adeos, Hyper-V, VMWare – VSphere and VCloud and Xen  Ganeti is a Virtual Machine cluster management tool developed by Google  uses either Xen or KVM as the virtualization platform

High Performance Computing  Computers are expected to process on large data in a shorter time  tradeoffs in achieving the desired process with the available resource  done with Super Computers  Examples:  3D Nuclear Test Simulation,  Genomic Assembly pipelines,  Satellite Image Processing

High Performance Computing (Cont.)  Can be achieved with Recent technologies  Grid Computing  Cluster Computing Systems  Cloud Computing  Our Focus is on:  Cloud Computing – Virtualized Environments

High Performance Computing (Cont.)  Challenges for High Performance Computing  lack of high speed interconnects  maximum of 10Gigabit Ethernet  Security of private data on Cloud Environments  Amazon Cloud Service  On-demand instances,  Reserved instances and  Spot instances

XEN – Hypervisor  Type 1 Hypervisor  Para Virtualization  Modules:  DOM0  Host Operating System  has complete access to hardware  DOMU  Virtual Machine which runs its own Operating Systems  Guest Machines  Hypervisor  Virtual Machine Monitor  complete access to control and manage CPUs, Memory

XEN Architecture:

XEN Architecture (Cont.)  access to the tables are restricted to the virtual machines  read access to the page tables  VMM is the key component  more levels of access  Handles traps and interrupts  latest development  XenoLinux  capable of running more than 100 virtual machines

HPC on Virtualized Environments  process large volume of data requires extensive computational power  to execute the set of instructions  produce results for the same  Hardware failures can be completely taken care  Example: For example, if one of HPC application is trying to run an intense algorithms and a Hardware failure is detected where the virtual machine is residing, the VMM can effectively migrate the virtual machine to the next available server without interrupting the tasks which are running in the virtual environment.

HPC on Virtualized Environments (Cont.)  increases productivity in HPC applications  VMM authorizes other virtual machines to get information  performance of the HPC applications benefits New Release Virtualization Spectrum

Conclusion  Virtualization concepts,  VM based computing in High Performance Computing application and  Xen Architecture and its applications  key notes in High Performance Computing  how High Performance Computing applications can benefits from the Virtualized environment  key features like Live Migration  Server consolidation would be the tedious task  saved by running the HPC in the virtualized environment

Thank You!