15
LINUX CLUSTERING WORKSHOP 4 Dr. Zahid Anwar

L inux clustering Workshop 4

  • Upload
    livia

  • View
    51

  • Download
    0

Embed Size (px)

DESCRIPTION

L inux clustering Workshop 4. Dr. Zahid Anwar. Simplified Architecture of Linux Cluster. Simplified Architecture of a Single Computer Simplified architecture of an enterprise cluster. No Single Point of Failure. An enterprise cluster should always have the following characteristic: - PowerPoint PPT Presentation

Citation preview

Page 1: L inux clustering  Workshop 4

LINUX CLUSTERING WORKSHOP 4Dr. Zahid Anwar

Page 2: L inux clustering  Workshop 4

Simplified Architecture of Linux Cluster

Simplified Architecture of a Single Computer

Simplified architecture of an enterprise cluster

Load Balancer

Cluster Nodes

Shared Storage

Print Server

Page 3: L inux clustering  Workshop 4

No Single Point of FailureAn enterprise cluster should always have the following characteristic:

“Any computer within the cluster, or any computer the cluster depends upon for normal operation, can be rebooted without rebooting the entire cluster.”

e.g. by building high-availability server pairs

Page 4: L inux clustering  Workshop 4

Clustering Terminology When a program runs When Process runs on a

Linux System A demon and the effects

it produces A service when combined

with its operating environment (config files, data, network)

When a resource moves from one computer to another

A proper failover configuration has no single point of failure.

Process

Service

daemon

Fail-Over

High Availability

Page 5: L inux clustering  Workshop 4

Types of ClustersOriginally, "clusters" and "high-performance computing" were synonymous.

Today, the meaning of the word "cluster" has expanded beyond high-performance to include

• high-availability (HA) clusters and

• load-balancing (LB) clusters

Page 6: L inux clustering  Workshop 4

Types of Clusters

High-availability clusters, also called failover clusters, used in mission-critical applications. The key to high availability is redundancy.

Load-balancing cluster provide better performance by dividing the

work This might be accomplished using a simple

round-robin algorithm. For example, Round-Robin DNS

Page 7: L inux clustering  Workshop 4

Cluster Computing?

Parallel

Computing

Distributed

Computing

Grid Computing

Cloud Computing

Page 8: L inux clustering  Workshop 4

Terminology Parallel computing

Tightly coupled sets of computation. E.g. Several pieces of data are being processed simultaneously in the

same CPU Homogenous collection of computers

Distributed computing Computing that spans multiple machines or multiple locations. Heterogeneous collection

Cluster Computing A form of Distributed Computing Generally restricted to computers on the same subnetwork or LAN.

Grid computing Frequently describes computers working together across a WAN or the

Internet. Much larger scale, tend to be used more asynchronously, and have much greater access, authorization, accounting, and security

concerns. Peer-to-Peer

Data or file-sharing (Napster, Gnutella, or Kazaa) SETI@Home

Page 9: L inux clustering  Workshop 4

Building a HA Cluster using Heartbeat

Heartbeat: ability to failover a resource from one computer to another

Functioning Tell Heartbeat which computer owns a particular

resource (define primary and backup server Heartbeat daemon on backup server listens to

the "heartbeats" coming from the primary server.

If backup server does not hear the primary's heartbeat, it initiates a failover and takes ownership of the resource.

Page 10: L inux clustering  Workshop 4

The Physical Paths of Heartbeat Normally Heartbeat configured to work over a separate physical connection between two servers.Separate physical connection can be either a

serial cable or another Ethernet

network connection (via a crossover cable or mini hub).

Adds extra traffic to your network

Page 11: L inux clustering  Workshop 4

Heartbeat Control Messages 3 basic kinds Heartbeat (status msgs)

Typically 150 bytes broadcast, unicast, or multicast

Cluster Transition msgs relatively rare contains conversation b/w daemons to move resources ip-request : to release the resource of ownership ip-request-resp :shuts off the service and no longer

owns the resource. On receiving ip-request-resp, it starts up the service and

offers it to client Retransmission Requests

Rexmit-a request for a retransmission of a heartbeat control message when one of the servers notices that it is receiving heartbeat control messages out of sequence.

Page 12: L inux clustering  Workshop 4

Secondary IP Addresses (Virtual Ips) Method for adding multiple IP addresses to the same physical network card.When you use Heartbeat to offer services it is done using secondary IP addresses

Page 13: L inux clustering  Workshop 4

Lab Exercise Set up a 2-node clusterConfigure a highly-available web server

Page 14: L inux clustering  Workshop 4

Load Balancing using Ultra Monkey (LVS)

Linux Virtual Server  (LVS) enables TCP/UDP connections to be load balancedMechanism of connection control is referred to Layer 4 Switching. Layer 3 IP address/port information is used.The host that LVS runs on is referred to as the Linux-Director (specialized router)Packets received for a virtual service by linux-director, routed by a scheduling algo

subsequent packets for the same connection sent to the same real server

Advantage of load balancer over round robin DNS

directs requests to less load nodes

accounts for sessions. (e.g. forum software, shopping carts)

Page 15: L inux clustering  Workshop 4

Overall mission

General Architecture

OS and Cluster Software

Cluster Hardware

Cluster Planning