Server consolidation Using Virtualization
Tony Thibault Sr. Technologist PPL Services ISD/ETS – Server Design
1
Reducing Costs Through Server Consolidation
In the past
Every application was installed on a dedicated standalone physical server. Server count was approaching 900 windows servers. Server’s were underutilized (average CPU utilization less than 10%). Many physical servers approaching EOL (ongoing issue). Data Center costs (cooling, power, space, cabling, power circuits, network equipment) rising and limited. New server deployments took 6 weeks (procurements, installations, setup, etc.)
2
Reducing Costs Through Server Consolidation
Moving forward:
No longer one application per physical server
Consolidation Strategies:
Application stacking – multiple applications sharing a single server (one operating system instance). This can be problematic from billing perspective, scheduling outages, potential s/w conflicts, vendor support. Server Virtualization – multiple operating system instances sharing physical server resources. Application Retirement – application no longer necessary.
3
Reducing Costs Through Server Consolidation
What is a Virtual Machine?
A self-contained software operating environment that behaves as if it is a separate distinct computer system. Each Virtual Machine appears to run on dedicated CPU(s), Memory, Disk Storage and Network Interfaces. The O/S and applications can't tell the difference between operating in a Virtual Machine or in a "real/physical" machine.
4
VMware Product Line
•
VMware workstation – installs as an application on top of Windows or Linux. (productivity tool for the user of one computer)
VMware GSX – installs as an application on top of Windows or Linux. Tuned to support more virtual machines and added support for remote console and scripting. (good for departmental use) Both options above incur the full overhead of a general-purpose operating system between the virtual machines and physical hardware. (supports limited number of virtual machines)
•
•
VMware ESX – installed directly on the bare metal. Higher performance. Supports more virtual machines per physical CPU than GSX. Advanced resource management. (datacenter deployments)
VMware Virtual Center – Centrally manage virtual infrastructure.
•
5
VMware product architectures
Workstation / GSX hosted on top of a generalpurpose operating system
ESX installs directly on baremetal. Developed specifically for hosting virtual machines
V M
V M
V M V M V M V M
Workstation/ GSX Windows/Linux x86 Hardware
Service Consol
VMKernel x86 Hardware
6
VMware ESX server under the covers
Physical NICs x 2
Physical Switch
VCONSOL
Physical Disks
Physical NIC x2 VMotion
Physical CPUs
Virtual Switch
VMKERNEL
Virtual Machine
Physical Memory
PPLnet
Virtual Switch
Physical NIC's 2 or more Virtual Machine
Virtual Machine
Virtual Machine Physical Switch Virtual Machine Virtual Physical Revision 1 - 11/18/2005 Tony Thibault Physical
7
VMware ESX at PPL implementation
VMware Production - Initial Design
DRAC
Vconsol (redundant connections - multiple switches
Fibre - seperate switch from other Fibre - seperate switch from other
VMotion (redundant connections - mulitple switches)
VMkernel - VM's seperate switch from other VMkernel - VM's seperate switch from other
P P L N E T
VMware Host
SAN
S A N
Fibre - seperate switch from other
DRAC Vconsol VMotion VMkernel - VM's seperate switch from other VMkernel - VM's seperate switch from other
PPLnet DRAC
Datacenter 1
F A B R I C
Fibre - seperate switch from other
Virtual Center Management Console
VMware Host
Datacenter 2
DRAC Vconsol Fibre - seperate switch from other Fibre - seperate switch from other SAN VMotion VMkernel - VM's seperate switch from other VMkernel - VM's seperate switch from other
VMware Host
NIC Speeds: DRAC - 10/100MB Vconsol - 1GB Vmotion - 1GB VMkernel NICS - 1GB Revision 1 - 11/18/2005 Tony Thibault
8
Advantages of Virtual servers
– Encapsulation – virtual machine data all contained in “files” called virtual disks. • Portability and Hardware Independence – the virtual disks can be moved/copied and then executed on most x86 compatible platforms - not locked into specific hardware platform or vendor (no more EOL servers). – On-the-fly resource management (virtual machine priority can be adjusted, additional compute resources (cpu, memory, disk, network) can be easily be added/removed. – Faster server deployments (no server procurement and delivery time, no physical racking, no pulling additional power circuits or network cabling, no delay in getting network ports and switches assigned, OS imaging from templates). – Windows O/S licensing costs reduced using Windows Server Datacenter licensing – Centralized management of virtual environment via Virtual Center – Optimal use of physical server resources (we are experiencing a server consolidation ratio of more than 20:1) – Server deployments in hours
9
VMotion and DRS
VMotion Technology
VMotion lets you move live virtual machines from one host to another while maintaining continuous service availability. This is useful if a host server needs to be repaired or upgraded. In addition ESX servers can be configured to dynamically load balance virtual machines across multiple ESX servers. This dynamic load balancing technology is called - DRS (Distributed Resource Scheduling).
10
VMotion – VM1 currently executing on VMware Host-A
PPLnet
167.155.22.xxx
VMware Host-A
VMware Host-B
SAN Fabric
Virtual machine is encapsulated into a disk file on the SAN
Virtual Machine 1 Disk File
Virtual Machine 2 Disk File
SAN (Storage Area Network)
11
VMotion – VM1 now currently executing on VMware Host-B without any disruption in service
PPLnet
167.155.22.xxx
VMware Host-A
VMware Host-B
SAN Fabric
Virtual machine is encapsulated into a disk file on the SAN
Virtual Machine 1 Disk File
Virtual Machine 2 Disk File
SAN (Storage Area Network)
12
High Availability
VMware supports a technology they call HA or High Availability. This technology allows for all Virtual Machines that had been running on a failed ESX server to automatically reboot on surviving ESX servers. DRS (Dynamic Resource Scheduling) can then dynamically balance the workloads across the surviving servers until the broken ESX server is repaired.
13
Questions
?
14