j hardware virtualization support in intel, amd and ibm power processors

Download j Hardware Virtualization Support In INTEL, AMD And IBM Power Processors

Post on 30-May-2018




0 download

Embed Size (px)


  • 8/14/2019 j Hardware Virtualization Support In INTEL, AMD And IBM Power Processors


  • 8/14/2019 j Hardware Virtualization Support In INTEL, AMD And IBM Power Processors


    support, simplicity, monitoring facility and security whichis the point of interest for indus trial computing systems.


    In uniprocessor system, it often assumes only oneprocess in the kernel. As a result, it simplifies the kernelinstructions and cross-process lock is not required. But thescenario is changed when multiple processors execute in thekernel. That means adding SMP support changes theoriginal operating system. Hence mechanisms forsupporting multiprocessors operating systems are required.There are different ways of organizing a multiprocessoroperating system such as giant locking, coarse-grainedlocking, fine-grained locking, asymmetric approaches,virtualization and API/ABI compatibility andreimplementa tion. But the virtualization technique is theimportant one as the developers are continuously upgradingthis technology . At first, we describe software-onlyvirtualization and hardware virtualization. Thenparavirtualization and full virtualization is explained.


    In software-only virtualization technique, the concept of 2-bit privilege level is used: using 0 for most privilegedsoftware and 3 for least privileged those. In this architecture(IA-32 and Itanium), the guest operating systems eachcommunicates with the hardware through the VirtualMachine Monitor (VMM) which must decide that access forall virtual machines on the system. Thus, the virtual machinecan be run on non-privileged mode i.e. non-privilegedinstructions can be executed directly without involving theVMM. But there are some problems that arise in software-only solution. Firstly, ring aliasing- problems that arisewhen software is run at a privilege level other than thelevel for which it was written. Secondly, address-spacecompression- occurs when guest software tries to access theVMMs guests virtual address space. Thirdly, impacts onguest transitions- may cause a tran sition to the VMM and notto the guest operating system. VMMs also face othertechnical challenges such as use of private memory forVMM use only, use of VMM interrupt handling, hiddenstate access etc. [16].


    Hardware virtualization allows the VMM to run virtualma chines in an isolated and protected environment. It is also

    transparent to the software running in the virtual machine,which thinks that it is in exclusive control of the hardware.In 1999, VMware introduced the hosted VMM, and it wascapa ble of extending a modern operating system to supporta vir tual machine that acts and runs like the hardware levelVMM of old [14]. To address the problems of softwareonly virtualization solution, hardware virtualizationmecha nism is applied which is possibly the most commonlyknown technology, including products from VMware andMicrosofts Virtual Server. Now, VMMs could run off-the-

    shelf operating systems and applications without recourse tobinary translation or paravirtualization. This capabilitygreatly facili tates the deployment of VMMs and providesgreater reliability and manageability of guest operatingsystems and applications.


    Basically, to overcome the virtualization challengesof software-only virtualization, the VMM was developedby the designers that modify guest software (source orbinary). Denali and Xen are examples of VMMs that usesource level modifications in a technique calledparavirtu alization. Paravirtualization is similar tohardware emula tion because in concept it is designed tosupport multiple OSs. The only implementation of thistechnology today is the Xen open source project, soon tobe followed by an actual product from XenSource.Paravirtualization provides high performance andeliminates the changes to guest ap plications. But thedisadvantage is that it supports lim ited numbers of operating systems. For example, Xen cannot support an

    operating system that its developers have not modified, suchas Microsoft Windows.


    Full system virtualization provides a virtual replica of thesystems hardware so that operating systems and softwaremay run on the virtual hardware exactly as they would onthe original hardware [13]. The first introduced software forfull virtualization system was CP-67, designed as a specializedtime-sharing system which exposed to each user acomplete virtual System/360 computer. Though fullvirtualization on PC architectures is extremely complex, atpresent it is pio neered in the market since 1998 as VMware

    initiated x86 based virtualization providing the fundamentaltechnology for all leading x86-based hardware suppliers. Itcreates a uniform hardware image that implemented throughsoftware on which both operating system and applicationprograms can run.


    The challenges imposed on IT business that the CIOsand IT managers always face are cost-effective utilizationof IT in frastructure and flexibility in adapting toorganizational changes. Hence, virtualization is a

    fundamental technological innovation that provides theskilled IT professionals to orga nize creative solutions tothose business challenges. The leading companies of ITsector are also introducing their innova tive and well-developed approaches every day to cope with demands of the age. Again the hardware virtualization support is animportant factor for the field of Grid Computing or se cureon-Demand Cluster computing. The hardware support forvirtualization in current microprocessors is addressed in thissection.

    (IJCSIS) International Journal of Computer Science and Information SecurityVol. 4, No. 1 & 2, 2009

    ISSN 1947 550073

  • 8/14/2019 j Hardware Virtualization Support In INTEL, AMD And IBM Power Processors



    Intel is developing microprocessors with variousadvanced virtualization supports. They are updating theirtechnologies constantly to facilitate the users demands.Starting with Server and mainframe systems virtualization,now Intel is providing hardware support for processorvirtualization through virtual machine monitor softwarewhich is also known as hypervisor. The actual aim of usinghypervisor is to arbitrate access to the underlying physicalhost systems resources so that multiple operating systemsthat are guests to VMM, can share them. IA-32 andItanium architecture were built on software-onlyvirtualization sup port [16]. But unfortunately they facedmany challenges while providing virtualization supports. Thesoftware cannot work properly in concern with the corehardware, thats why it has to use complex schemes toimitate hardware features to the software. Moreover, it hasto make the illusion that the host operating system thinkingthe virtual machine as another application. To eliminate theseproblems, VMM designers developed new solutions likeXen [2] and Denali VMMs that use source levelmodification known as paravirtualization. But the mainlimita tion of this scheme is that it is applicable for a certainnumber of operating system. Hence Intel developed newarchitectures VT-x and VT-i for IA-32 processors (CoreDuo and Solo) and Itanium processors family respectivelywhich offered full virtualization using the hypervisorsupport. This new architecture enables VMM to run off-the-self operating systems and applications without any binarytranslation or paravirtualization. As a result it in creasesrobustness, reliability and security.


    AMD has introduced their new Quad-Core AMD OpteronProcessor (based on Pacifica specification) which is designedto provide optimal virtualization. This processor provides anumber of features which enhances the performance andefficiency of the virtualization support. Firstly, AMD OpteronRapid Virtualization Indexing, which allows virtual machineto more directly manage memory to improve performance onmany virtualized applications [1]. It also decreases theworld-switch time i.e. time spent switching from one virtualmachine to another. Secondly, direct CPU-to-memory, CPU-to-I/O, and CPU-to-CPU connections to streamline servervirtualization is ensured through AMDs direct connectarchitecture. Thus, it is possible to host more VMs per serverand maximize the benefits of virtualization in terms of highbandwidth, low latency, and scalable access to memory.Thirdly, tagged Translation Look-Aside Buffer (TLB) hasincreased responsiveness in virtualized environments.Actually, through Tagged TLB, AMD Opteron processormaintains a mapping to the VMs individual memory spaceswhich eliminates additional memory management overheadand reduces switching time of virtual machines. Finally,Device Exclusion Vector (DEV) performs security checks inhardware rather than software. DEV mechanism controlsaccess to virtual machine memory based on permission. Theseunique features have brought AMD to the frontline of battleon hardware virtualization support.


    As the successor of POWER3 and POWER4, IBMintro duced advanced virtualization capabilities in IBMPOWER5 processors in 2004. This processor includesincreased per formance and other functional enhancements of virtualization- reliability, availability, and serviceability inboth hardware and software levels [9]. It uses hypervisorwhich is the basis of the IBM virtualization technologies onPowers systems. This technology provides fast page moverand simultaneous multithreading which finally extends thecapability of PPC5. It supports logical partitioning andmicro partitioning. Up to ten LPARs (logical partitions)can be created for each CPU. Thus the biggest 64-Waysystem is able to run 256 independ ent ope