1. The vMatrix: Server Switching IEEE FTDCS’2004 Amr A. Awadallah Mendel Rosenblum {aaa,mendel}@cs.stanford.edu Stanford University – Computer Systems Lab – IEEE FTDCS 2004

2. Static Servers (ala Circuit Switching) FINANCESPORTS Load Balancer

3. Stanford University – Computer Systems Lab – IEEE FTDCS 2004 Server Switching (ala Packet Switching) FINANCE & SPORTS Load Balancer vMatrix Oracle

4. Stanford University – Computer Systems Lab – IEEE FTDCS 2004 Server Drop (ala Packet Drop) FINANCE & SPORTS Load Balancer vMatrix Oracle

5. Stanford University – Computer Systems Lab – IEEE FTDCS 2004 Advantages of Switching: Hardware Savings from multiplexing services over servers. People Savings from managing less hardware. Better Scaling in cases of traffic spikes. Key: Peak of Sums < Sum of Peaks

6. Stanford University – Computer Systems Lab – IEEE FTDCS 2004 What makes Switching hard ? Software Dependencies Hardware Dependencies People Dependencies

7. Stanford University – Computer Systems Lab – IEEE FTDCS 2004 What is The vMatrix?

8. Stanford University – Computer Systems Lab – IEEE FTDCS 2004 Virtual Machine Monitors Virtual Machine Monitor Real Machine (CPU, Memory, Disks, Network) Virtual Machine 1: vCPU, vMem, vDisk, vNet OS1: Windows 2000 Virtual Machine 2: vCPU, vMem, vDisk, vNet OS2: Linux MySQL, Apache Oracle, IIS

9. Stanford University – Computer Systems Lab – IEEE FTDCS 2004 The vMatrix! A network for delivering virtual machines (VMs) between real machines (RMs) running the virtual machine monitor (VMM) software.

10. Stanford University – Computer Systems Lab – IEEE FTDCS 2004 Advantages Presents developers with same isolation model Backward Compatibility Efficient Resource Utilization On-Demand Spike Absorption Faster Recovery Disadvantage VM files are very large (order of gigabytes)! VM Software is not free

11. Stanford University – Computer Systems Lab – IEEE FTDCS 2004 vMatrix Framework: The Oracle: This is the main control program RM VMM VM Hibernation Nest: VM files are suspended and stored here for later retrieval. Production Cluster: Live operational VMs, typically only one VM per RM Loading Chambers: Development and Maintenance Cluster, there could be more than one VM per RM here. VM RM VMM VM VM VM RM VMM VM VM VM RM VMM VM RM VMM VM

12. Stanford University – Computer Systems Lab – IEEE FTDCS 2004 Lifecycle of a server VM Loading Chambers Hibernation Nest Instantiated 1. A large number of virtual servers are stored as dormant files on a SAN or Netapps 2. A number of virtual servers are activated in a shared RM so that developers & system admins can upgrade, debug, or patch them. 3. A virtual server is activated on a dedicated RM and exposed to live load.

13. Stanford University – Computer Systems Lab – IEEE FTDCS 2004 Experiences: Real Machine (PIII-550MHz, 640MB RAM, 9GB hard disk) VMware ESX VMM Server (consumes 184MB RAM, 3.5GB hard disk and 5% CPU) Virtual Machine exposes a PIII-550MHz with 512MB RAM and 5.5GB hard disk. Operating System: Red Hat Linux 9 Apache Web ServerMySQL Database PHP (Hyper Text Processor) PHP-Nuke and osCommerce Internet Services Real Machine (PIII-550MHz, 640MB RAM, 9GB hard disk) VMware ESX VMM Server (consumes 184MB RAM, 3.5GB hard disk and 5% CPU) Virtual Machine exposes a PIII-550MHz with 1024MB RAM and 5.5GB hard disk. Operating System: Yahoo FreeBSD 4.8 YLIB (Yahoo C/C++ Libraries) YSS (Yahoo Structured Search) Yahoo! Autos Search Network API

14. Stanford University – Computer Systems Lab – IEEE FTDCS 2004 Previous and Related Work Common drawback: no backward compatibility. Application Servers: IBM WebSphere, ATG Dynamo, BEA WebLogic, JBoss. (no strict isolation, interface violations). Java Virtual Machine. (low performance). Packagers: Debian APT, Redhat RPM, PRC. (no isolation). Disk Imaging / Ghosting /Chrooting. (no concurrent VMs). Light Weight OSes (Denali, Xenoservers). (new OS). OS Virtualization (Ensim, Ejasent, EXETender). (tied to OS).

15. Stanford University – Computer Systems Lab – IEEE FTDCS 2004 Conclusion Server Switching solves many problems: (1) Better Resource Utilization (2) Peek Absorption The vMatrix is a practical backward- compatible solution for alleviating the people/software/hardware dependencies of Server Switching.