Virtualisierung für Linux, Windows und Solaris. Thomas Finnern DESY/IT-SysOps

Transkript

1 Virtualisierung für Linux, Windows und Solaris Thomas Finnern DESY/IT-SysOps

2 Agenda Eine Einführung in die Thematik Vergleich von XEN und Solaris Containern Ein bisschen Windows Stand und Ausblick für unser Rechenzentrum Live-Vorführung von Christoph und Martin 2

3 Allgemeiner Einstieg 3

4 Overview Virtualization Why Features How Hypervisor Guests Use Cases Where to Conclusions 4

5 Definition Network Resources Storage Resources Used Virtualizations Clustering Sun Cluster Content Based Routing F5 Layer 7 Routing Cisco? Host Virtualization Solaris Container XEN Linux Windows Virtualization Virtualization is the technique of managing and presenting storage devices and other resources functionally, regardless of their physical layout or location. Hypervisor Root User App OS OS In computing, paravirtualization is a virtualization technique that presents the abstraction of virtual machines with a software interface that is similar but not identical to that of the underlying hardware. 5

6 Why Minimize Efforts Easy Provisioning Easy Resource Control Multi-OS Service Offer Security / Service Separation Getting Better Enhanced Load Distribution Enhancing Fault Tolerance and Security Separate Test, Developing and Production on same Hardware 6

7 How RedHat Start: RHEL 5.0 with XEN 3.0 Plan: RHEL 4.5+, 5.0+ with Built-In Support including AFS Logical Volume Manager Hypervisor and Guests Sun Solaris 10 Zones Plan: (Open)Solaris 11 with XEN ZFS (Solaris 10 U3+) Zone + SRM = Container 7

8 Key Features and Benefits XEN: Open Source Complete Separation between Guests Paravirtualization Intel-Vanderpool and AMD-Pacifica Supported for unmodified Guests (without Live Migration) Virtual Machines near Native Speed Live Migration of Guests Up to 32 virtual CPU s per Guest Load Distribution and Weighting between Guests X86/32 and x86/64 Support Network over Software Bridge(s) Own Mac-Address for each Interface Solaris Container: Open Solaris Separation (and Shares) configurable One OS for All Sparc and x86 Guests Machines at Native Speed No Live Migration (?, cloning planned) Cores per Zone not limited Resource Management between Containers Sparc and x86 with 32/64 bit Support Network over selected Device(s) No own Mac-Address for each Device (?) Sparse-root and Whole-root Models 8

9 Hypervisor / Global Zone 9 Customisation XEN Test: Defined one admin domain cross platform Configuration: Add feature vms Reconfiguration: New Install / Reboot Hypervisor with Minimum Installation Logical Volume Manager for FS-Handling Separate Network for Service and Production Guest Installations from Scratch/Image? Unvisible Filesystems /scratch/vsges/root /scratch/vsges/sge /scratch/vsges/ Changing to Image Preparation Image Preparation Partitioning OS and Data Application Networking Deployment Methods mk_image mk_virtual Live Cycle Management As Usual Methods mk_image Install File or Clone/Copy Pack mk_virtual Unpack Networking Partitioning Glue Solaris Container Test: Defined one admin domain cross platform Configuration: Add feature vms Reconfiguration: On the fly Global Zone Installation will/can be inherited ZFS for FS-Handling Separate Network for Service and Production Guest Installations from Scratch/Image Visible Filesysstems /scratch/vinnetou/root /scratch/vinnetou/afscache /scratch/vinnetou/

10 Guests / Zones and Pools XEN Test: Configuration: Add feature vms-<server> Openafs kernel module must be compiled XEN-Mac 0x00FFFFFF&IP 0x (Change to XEN mac=00:16:3e:xx:xx:xx?) Solaris Container Test: Configuration: Add feature vms-<server> afs can be inherited (ro?) Next version: privilege grant for kernel module usage Slight differences in SVC (Service Management Facility) for own sshd startup Testing NIS-Setup 10

11 Expected Use Cases Separation of Applications and Services Providing Service IP Addresses Application Specific User Registry and System Settings License Cost Optimisation e.g. Oracle Capped Containers Application Specific Resource Binding Storage Network CPU Peak Load Elimination Multi OS Offer for Customers Test Environments Developing Environments Migration Support Multi Customer Pools e.g. SGE Automatic Deployment of e.g. WEB- Services Educational Environments for Admins and Students Database Factory Security by Encapsulation New Service Environments Order and Delivery within 1 Hour Central Image Handling over SAN Ease Computer Adminstration... New Operating Concepts Installation Testing Monitoring Backup Patching 11

12 Where to Server Consolidation (Cost, Heat, Admin) Running Solaris and Linux on one Host Running 100-Core Hosts over Internal Crossbar as Farm or WEB Appliance Cross Cluster Hopping Live Migration Image Moving Configuration and Life Cycle Management Buy xx Open Xen Licences 12

13 Conclusions Virtualization Is Simple To Use Nice Operating and Support Model Administrate only Big Boxes Cross Platform Support Future Real or Para Virtualization will be standard XEN Hypervisor for Solaris and vice versa Standard provisioning frame (kernel, partition, image, data, ) Security Don t let your system be virtualized by somebody else 13

14 IT-SysOps: WBOOM-Configuration Guest knows Hypervisor(-Cluster): SF=vms- <CHOST> Hypervisor knows all guests: woom:vms BIRD Cluster: Common Store in AFS Queue and Shadow Master (it-services01/2) 2 Host Login Cluster(it-services01/2) 16 Exec Cores (bird009/bird010) Library spires-container (spires) WOF Test Apache (vinnetou,vinnern) Plan: Netra-Consolidierung Netapp for Live-Migration Yves Kemp: Im Tier2-Batch noch nicht 1 OS SL3 -> SL4 Sven Sternberger: Seit ca. 2 Jahren haben wir mit der Whithouse einen Server der den Entwicklern schnell neue SL3 bzw SL4 Installationen zur Verfuegung stellt. Sie habe die Moeglichkeit hier neue Software zu installieren und verschiedene Versionen von Installationen zu sichern. Durch die Nutzung von rsync ist es moeglich in wenigen Minuten zwischen verschiedenen Setups zu wechseln, und so die komplexen Setups aufzubauen die fuer eine Softwareentwicklung von einem verteilten System wie dcache notwendig ist. Owen Synge: Seit ca. 1 Jahr entwickelt Owen an einem System zur Automatisierung der build und test Umgebung. Das bedeutet ein Entwickler checkt eine neue Version ein in der Versionsverwaltung. Daraufhin wird automatisch eine frische SL3 bzw. SL4 Installation generiert, die build Umgebung wird installiert und aus den neuen Sourcen ein aktuelles Packet gebaut. Im folgenden werden mehrere Instanzen generiert, die mit der neuen Software ausgestattet werden und mehrere Testsequenzen werden abgefahren um die Funktionsfaehigkeit des neuen Releases sicherzustellen. 14

15 Live Demo Christopher Open Solaris zoneadm list -vc zlogin GUEST ps -efz reboot Martin Red XEN Linux xm list xm console GUEST xm top reboot 15

16 Notizen 16