prezentációt

University of

Michigan Administrative Information Services

Server Virtualization Technologies: Uses, Comparisons, and Implications

David Sweetman Windows Enterprise Systems AdminAdministrative Information Services

University of Michigan

[email protected]

Spring, 2005 Windows Virtualization Technologies 2

Presentation Overview

• The What and Why of virtualization

• Comparing Product Features

• Comparing Product Performance

• Evaluating Physical Servers for virtualization

• Costs

• Questions


What is server virtualization?

• Creating multiple logical server OS instances on one physical piece of hardware

• All HW drivers are virtualized – same virtual HW regardless of physical HW

• Each virtual machine is completely independent of the others and doesn’t ‘realize’ it’s virtualized


Why virtualize?

• More efficient HW utilization

• More efficient staff

• Long-term matching resources & needs

• Quick and nimble server provisioning

• Testing & Troubleshooting

• More effective redundancy

• HW maintenance w/o app downtime

• Simplify system imaging

• Disaster Recovery


Individual ebb and flow of resources

Cumulative usage of 28 servers in the MAIS data center evaluated for virtualization:

44GB RAM, 138.15Ghz CPU, and 1323GB HD

45% of RAM not used 99.9% of time.

25% of RAM never used concurrently.85% of CPU not used 99.9% of time.

81% of CPU never used concurrently.

HW Utilization Facts

68% of hard disk space unused


Hard Disk UtilizationServer Local Disk Total

(GB)Used (GB)

Free (GB)

SAN Manager 48 8 40

IIS app test 68 9 59

TNG Scheduling 68 13 55

PeopleSoft 8 HE 34 7 27

PeopelSoft 8 FIN 34 24 10

IIS / SQL:Research app 68 31 37

Small use Citrix 17 9 8

File Servers 136 56 80

Stat Version Control 34 6 28

Stat Version Control 17 6 11

SQL: eLearning dev 68 16 52

IIS: eLearning dev 68 11 57

SQL: eLearning Prod 68 10 58

IIS: eLearning Prod 34 13 21

Machine Room environ 68 6 62

IIS document server 170 88 82

Domain Controller 34 7 27

More Efficient Hard Disk Utilization Total: 1323 GB Used: 418 GB Free: 905 GB (68% unused)

SAN in 30GB chunks

1 fibre channel >1 server

Virtual HDs more granular

Share free space – allocate as needed


Virtualization vs. Consolidation

• Virtualized servers = separate OSes

• Consolidation = same OS

• Virtualized servers must each be administered, patched, etc.

• Consolidated applications can introduce conflicts and support issues


Virtual Host Licensing

Windows and other Microsoft per-server apps are licensed per virtual server. (1 physical server w/ 6 virtual Windows servers = 6-7 licenses needed)

As of 4/1/2005, Microsoft per-processor licenses are per physical processor (1 physical server w/ 3 virtual SQL Servers sharing 1 CPU = 1 per-processor license)

Virtualization savings are not in licenses.Check with other vendors.


Virtualization Software

• MS Virtual PC 2004 – workstation only

• VMWare Workstation 5 – workstation only

• MS Virtual Server 2005, Standard (4p)

• MS Virtual Server 2005, Enterprise (32p)

• VMWare GSX Server 3.1

• VMWare ESX Server 2.5


Common Features

• Up to 3.6GB RAM per virtual host• Web-based console for administration• Host OS sees HT CPU, virtual do not• VMs consist of 1 config file & 1 file / HD• VMs can mount physical CDs or ISOs• VMs can be multi-homed• Up to 64 VMs per host server• Highly scriptable – extensive API• Granular permissions for individual VMs• Detailed logging


MS Virtual Server 2005

• Targeted to increase efficiency in testing and development, and “re-hosting”

• Up to 1 processor per virtual host

• Windows = underlying host OS

• Only Windows VM’s supported

• No USB support

• 2 processor SMP coming soon


VMWare ESX Server 2.5

• Targeted at mission-critical enterprise services

• Up to 2 processors per host

• Custom Linux = underlying OS

• Windows & Linux VM’s supported

• Dedicated NIC for admin (2 total min)

• USB support

• 4 proc SMP coming soon


Do I need to know Linux?

• VMWare ESX Server is based on Linux

• All administration is possible through web

• Don’t need any Linux experience for installation or ongoing admin

• SSH and SFTP access to server

• Used? Installed backup software sFTP’ed ISO’s to server


Managing Virtual Servers

• Web site is primary interface• Attach to VM console

Virtual Server = ActiveX control VMWare = separate application

• Reboot, power on, power off• Create and manage VM’s• Allocate hardware resources• Mount CDs and floppies• View recent performance data


VS Screenshot


VMWare Screenshot


Hyper-threading

• One physical CPU seen as 2 logical

• Both products see HT, non-HT VMs

• Slows virtualization performance

• 1 HT CPU < 2 Phy CPU

• 0-20% performance increase over no HT

• http://www.intel.com/technology/hyperthread/


RAM Allocation

• Virtual Server: Max <= total physical memory

• VMWare: Max <> total physical RAM Ballooning RAM pooled across multiple VMs Enables more efficient RAM utilization If max out, goes to paging file


VS Screenshot


VMWare Screenshot


Monitoring

• MOM (or other host monitoring): Monitors VMs like physical

• Virtual Server: MOM Management Pack Integrates into MOM framework Monitor overall host and VM servers

• VMWare: vmkusage

• VMWare: VirtualCenter Database back-end across all servers


Virtual Center

• Central monitoring and management in VMWare environment

• Manage all VMs from one interface

• Additional software / license

• Management application

• Set thresholds and actions – like MOM

• SQL or Oracle DB backend

• Assign privileges via NTFS


Virtual Center Screenshot


Converting Physical Server

• Both MS & VMWare offer tools to create virtual systems from physical

• Physical HW drivers replaced by VM• Ideal for the truly unique server (highly customized)• Both vendors recommend loading virtual servers from

scratch• Slow for both vendors – 6h / 4GB image• VSMT (Virtual Server Migration Tool)

many prereqs (DHCP, ADS, SQL) Not in one month eval

• P2V (Physical 2 Virtual) Simple boot CD and ‘server’ piece Licensed per use


VMotion

• Enables seamless transition of live virtual host between physical servers

• Dynamic Resource Allocation across servers – respond to load changes

• HW maintenance


Best Practices

• Plan out server allocations

• Create “gold image” – base OS kept up-to-date patches – duplicate for new VMs

• Use ISO’s for CD access

• Use standard backup and restore

• Take system images as needed


Summary of VMWare differences

• More comprehensive web GUI (for example, deleting hosts & HDs)

• Support for dual processor virtuals

• Support for Linux virtuals

• Virtual Center: central management

• Easy-to-use physical-to-virtual support

• VMotion: seamlessly move virtual servers between physical hosts


Testing Environment

• One month each was spent evaluating MS Virtual Server & VMWare ESX Server

• Identical testing was attempted on each. Load and usability testing: Win 2000, 2003, IIS5, IIS6, SQL Server 2000, 3rd party apps

• Test hardware 1.4Ghz x 4 physical processors (8 w/ HT) 8GB of RAM 60GB fibre-channel connected SAN space


Performance Comparisons

• Automated load test of Aspen 2.5 dev environment (Win 2000/IIS5 & Win 2000/SQL 2000)

• Citrix / TS load test w/ Helpdesk• IIS6-based memory, CPU, disk, and network

I/O testing• SQL Server add, update, and delete testing• Load testing both as isolated server and with

other virtual server processing• ‘Normal usage’ w/o issue in all cases



Physical MSVS VMWare

CPU 100% 94% 80%

Memory 100% 91% 91%

Disk&NIC I/O 100% 101% 101%

SQL 100% 57% 87%

• Windows 2003 IIS6 and SQL 2000 perf compare

• VMWare CPU : hyper-threaded related, ~93% w/o

• VS SQL : VS 2005 SP1 has performance enhancements



• Previous stats were isolated tests

• VMs won’t be alone on physical host

• How does system perform w/ other VMs running assorted, intensive tasks?

RAM CPU Disk Network

Virtual Server 2005 -/+ <10% -/+ <10% - <5% - <5%

VMWare ESX Server Same Same - <5% - <5%


IIS/SQL Load Test Results

• Mercury LoadRunner scripted test

• Overall performance 100@30/min: VM = 60% 1000@12/min: VM = 99%

• What made it slow? CPU queuing Memory, HD, NetIO – nearly identical


Terminal Services / Citrix Load Test Results

Currently 14 servers, 4procs (8HT), 4GB RAM –load balancing ~700 concurrent

CPU and RAM intensive apps~60 users max per physical server

CPU = bottleneck (logon & BusObj)1CPU = 7 users max ; 2 CPU = 12 max100 v 1CPU or 58 v 2CPU to match 14 physicals

Recommendation: 2 CPU & only for small use


Business Objects WebI dev

Virtualize? Yes. 900 / 1.4Ghz

24 x 7 sampling Hours of Op sampling

RAM (MB)

Nic KB/sec

HD KB/sec

Proc % Usage

RAM (MB)

Nic KB/sec

HD KB/sec

Proc % Usage

100% 847 61 4324 112 847 61 4324 34

99.99% 839 45 1548 55 839 47 97 24

99.9% 823 19 487 52 821 32 85 22

99% 816 17 76 3 816 17 67 7

95% 813 4 65 3 814 11 65 6

90% 809 2 64 2 812 2 63 5

Av 755 2 55 2 759 2 52 1

StDev 47 4 86 3 44 4 136 1

Win 2000 / IIS5 / 2400MB RAM / 1.4Ghz x 2 (no HT)


PSoft 8 Fin Crystal/nVision: Dev

Virtualize? Yes. 900 / 1.4Ghz


RAM (MB)

Nic KB/sec

HD KB/sec

Proc % Usage

RAM (MB)

Nic KB/sec

HD KB/sec

Proc % Usage

100% 835 5284 5038 106 835 5284 2529 106

99.99% 822 2973 5005 89 829 4542 1555 92

99.9% 821 31 4912 82 822 72 1494 75

99% 814 4 4802 67 819 5 1015 13

95% 798 3 910 10 809 3 100 3

90% 786 3 61 2 799 3 32 2

Av 644 2 205 4 662 3 49 2

StDev 91 51 834 8 106 87 181 3

Win 2000 / 2300MB RAM / 1.1Ghz x 2 (no HT)


PSoft8 HE Crystal/nVision - Prod

Virtualize? NOT at this time – CPU needs too high


RAM (MB)

Nic KB/sec

HD KB/sec

Proc % Usage

RAM (MB)

Nic KB/sec

HD KB/sec

Proc % Usage

100% 716 11499 3421 350 716 9437 3421 348

99.99% 710 9803 3379 329 713 3493 3379 333

99.9% 620 1422 2440 244 691 674 779 267

99% 534 119 2304 192 460 119 351 220

95% 483 8 183 34 440 11 74 32

90% 447 1 50 23 437 1 51 22

Av 378 10 105 23 363 7 49 21

StDev 67 162 356 7 63 109 159 6

Win 2000 / 1500MB RAM / 2.8Ghz x 1 (w/ HT)


sumTotal Aspen 2.5 eLearning

Virtualize? Yes 2300MB / 1.4Ghz x 2 Note: high NIC=sync ; CPU=imp/exp


RAM (MB)

Nic KB/sec

HD KB/sec

Proc % Usage

RAM (MB)

Nic KB/sec

HD KB/sec

Proc % Usage

100% 2077 9061 4477 277 2077 406 1047 155

99.99% 2075 5865 3682 233 2075 404 1039 149

99.9% 2073 2667 3673 216 2073 206 971 138

99% 1984 91 3626 138 2064 70 827 125

95% 1777 68 839 101 1684 67 623 59

90% 1670 5 517 41 1665 3 459 30

Av 1628 16 236 24 1636 5 166 21

StDev 76 173 505 5 60 20 183 4

Win 2000 / SQL 2000 / 2358MB RAM / 1.9Ghz x 2 (w/ HT)


Domain Controllers

Virtualize? Yes – 850MB / 1.4Ghz


RAM (MB)

Nic KB/sec

HD KB/sec

Proc % Usage

RAM (MB)

Nic KB/sec

HD KB/sec

Proc % Usage

100% 776 5677 4298 146 767 457 1237 92

99.99% 771 5326 3674 131 766 98 195 51

99.9% 768 2131 3440 78 757 93 194 16

99% 753 51 1972 43 753 42 180 13

95% 713 24 140 12 713 27 90 11

90% 707 15 91 10 707 20 88 10

Av 633 12 128 8 646 7 78 8

StDev 74 138 302 3 56 11 41 1

Win 2003 / 2000MB RAM / 700Mhz x 4 (no HT)


Univ of Michigan - Flint

• VMWare ESX Server

• Determining factor: Linux support & MS Virtual Server wasn’t available

• Several years of experience, starting with GSX, public web services, online teaching, real video server, internal file/print, 46v on 5 physical (15 on 1), <10% slower, Dell 2650’s & 4600’s, 2 proc, 12GB RAM


NC State University

• MS Virtual Server 2005

• Determining factor: Cost

• PeopleSoft v8 Crystal/nVision app servers: 18 virtual servers, 7 physical servers, dual Xeon >2GB, physical v. virtual head-to-head, little difference in performance.


Potential Uses from Previous Presentations

• NAP - Remediation Servers – “Big Red Button” for critical fix – assign additional resources

• Keynote - Reliability – one of pillars of Trustworthy Computing

• Boston U – Matt - NetReg peak usage first couple weeks of semester

• WSUS 3Ghz, 1GB RAM recommended – sitting idle most of time?

• Decrease dev system allocation in busy times


Pricing

MS Virtual Server 2005 (4CPU Server, 8GB RAM) Win 2003 Std: up to 4 processors, Ent: up to 32 VS Std: 4proc/4GB; Ent: 8proc/32GB 2003 Ent/Std: ~$500+~$500 = ~$1000

VMWare Server ESX (4CPU – other pricing scales) ESX: $4500/phy server + $945/yr support ESX+SMP+V-agents: $6000/phy server

+ $1764/yr support

VMWare Add-ons VirtualCenter server: $3000 + $1050/yr P2V Starter kit (25): $2000 + $420/yr


Cost / Benefit Example

• VMWare Server ESX $45K separate HW purchase price $29K + $2K/yr (ESX w/SMP): ~35%

• MS Virtual Server Std $33K separate HW purchase price $30K virtual HW + software: ~10%

Note: In both cases, estimates are conservative


Summary / take-aways

• More effective resource utilization and response to changing needs (5-15% to 60-70%)

• Virtual Server & VMWare = comparable performance, VMWare more isolated

• VMWare more feature-rich: SMP, VMotion, manage multiple servers

• VMWare costs more, but you can do more, virtualize more costly servers

• Both platforms have limits, active improvement


Other Resources

• VMWare: www.vmware.com

• Virtual Server: www.microsoft.com/virtualserver/

• Rapid App: www.rapidapp.com

http://www.vmware.com/

http://www.microsoft.com/virtualserver/

http://www.rapidapp.com/


David SweetmanUniversity of [email protected]

Questions?

prezentációt

Documents

host server

server virtual hds

server apps

virtual hw

p ms virtual server

ram allocation virtual

physical server w

vmware esx server