Virtualization of Middleware Servers

thingsplaneΔιακομιστές

9 Δεκ 2013 (πριν από 3 χρόνια και 10 μήνες)

93 εμφανίσεις

Virtualization of Middleware Servers
Sven Gabriel
Nikhef
4.Feb.2008
Introduction
Motivation
Xen
Technology - Terminology
Integration in Site-Management framework
Summary and Next Steps
Server Virtualization
I
Situation:Different
services with different
hardware requirements.
I
Run more services in
one server:
Maintenance issues
I
Run each service in a
separate node:waist of
resources
Server Virtualization
I
Virtualization:
I
Services isolated,
easier maintenance
I
Dynamically adjust
resources (RAM,
CPU,HD)
I
Snapshots (lvm),
Cloning,Migration
I
Redundancy,same
service multiple
instances
I
Easy setup of
Test-sites
Server Virtualization
I
Goal:Integration of the
Virtualization Scheme in
the existing
Management
environment (kickstart,
Quattor)
Virtualization Ans¨atze
I
Virtualization on Operating System level -Process
Virtualization - processes of the virtual instances are
separated form each other -no separate kernels started.
I
Examples:OpenSolarisZones,BSD Jails,Linux VServer,
OpenVZ
I
Full Virtualization -all x86 based systems can be
Virtualized Guest Systems (run unmodified kernel) no
restriction to the host kernel as in Process Virtualization
I
Examples:VMWare Server,-GSX Server,Microsoft Virtual
PC;Aprox 20-25 loss in performance,(VMWare ESX
Performance is much better)
I
Hardware Based Virtualization Example:Logical
Partitioning (LPAR) IBM z and p series
I
Para Virtualization -Xen,modified Kernels,Hardware
access via hypervisor,very little performance losses
Paravirtualization
I
menu.lst:kernel
xen.gz-2.6.18-53.1.6.el5
starts Dom 0:vmlinuz-
2.6.18-53.1.6.el5xen
I
Dom 0 has direct access
to the HW(runs the xend
which is the interface to
the hypervisor
I
Dom U (unprivileged
domains have no direct
HW-access)
I
Split Drivers
Paravirtualization
I
menu.lst:kernel
xen.gz-2.6.18-53.1.6.el5
starts Dom 0:vmlinuz-
2.6.18-53.1.6.el5xen
I
Dom 0 has direct access
to the HW(runs the xend
which is the interface to
the hypervisor
I
Dom U (unprivileged
domains have no direct
HW-access)
I
Split Drivers
Full Virtualization
I
Hypercalls:extension of the
x86-Commands
I
Hypervisor runs in Ring 0,controls
HWaccess,passive vgl.
ESX-Server
I
Xenified Guest contact Hypervisor
for privileged (Ring 0) operations
I
all IO-Operations have to pass the
Hypervisor
I
Hypervisor can schedule,priories.
Guests can be weighted.
I
VT-x/VT-i;Pacifica Processors have
an extended command set
(Hypervisor)
I
Only privileged Domain gets direct
HW-Access,unprivileged are
directed to the Dom 0
Performance
I
Dom 0:Centos-5.1 Xen-3.1 i386 Dom U:Centos-4.6 i386
I
Network (iperf):HVM:< 60 Mbit/s PVM:900 Mbit/s
Performance
I
Dom 0:Centos-5.1 Xen-3.1 i386 Dom U:Centos-4.6 i386
I
Network (iperf):HVM:< 60 Mbit/s PVM:900 Mbit/s
What we can do now
I
Dom0:Centos-5.1 Xen-3.1 x86-64 (Kickstart Managed)
I
Possible Guests (DomUs):
I
Xen Administration Tools:Centos-5 (64,32),Centos-4
(64,32),Centos-3 (32,2.6 kernel)
I
pypxeboot/kickstart:Centos-5(64,32),Centos-4 (64)
I
Dom0:Centos-5.1 Xen-3.1 i386 (Kickstart Managed)
I
Xen Administration Tools:Centos-5 (32),Centos-4 (32),
Centos-3 (2.6 xenU-kernel)
I
pypxeboot/kickstart:Centos-5(32),Centos-4 (32,needed
for middleware services)
I
Quattor installation of a Top-Level-Bdii in Dom-U,needs
some handwork
Boot Concept
I
xm (Xen Manager) Admin tool sets besides others the boot
mechanism
I
kernel and initrd in dom 0/boot/*xenU
I
Or pygrub:uses the dom-U filesystem holding the GRUB
configuration
I
Or pypxeboot:needs MAC of Dom-U (udhcpc),gets pxecfg
and ks via tftp drops to pygrub if pxeboot is set to boot
I
Or (RH) virt-install,virt-manager installation/management
tools for dom-Us
I
Or...
Summary
I
Paravirtualized Guests show better performance
I
64 bit Dom-0 more flexible
I
Quattor installation in Xen-Dom-U works.(needs clean up)
I
Existing Monitoring Tools (Nagios,Ganglia) can be used
for Xen-Guests
I
Centos flavors 3.9 - 5.1 in 64 and 32 bit can in principle be
Virtualized
I
C-5.1-32 Dom-0s and C-4.6-32 Dom-Us can be
used/managed within the existing Installation/Configuration
framework
I
Next Steps:Clean up,Tests:run several virtualized MW
Services
I
Storage,and Cluster-Management