Implementing a Load-balancing Web Server Using Red Hat Cluster Suite

1
Implementing a Load-balancing Web Server Using
Red Hat Cluster Suite

• Ya-Lee Tsai

2
Project Goals

• Building a load-balancing web server
• Performance evaluation

3
Steps

• Setting up a working environment.
• Implementing load-balancing support in the
  system.
• Performance measurement and comparison.

4
Load-balancing Web Servers

• Balancing the load among the servers
• Providing
• High throughput
• High availability

5
Two Classes of Locally Distributed Architecture
for Web Sites

• Cluster-based web systems
• Virtual IP address
• Distributed web systems
• IP address visible to client applications

6
Architecture of A Cluster-based Web System
7
Architecture of a Distributed Web System
8
System Environment

• Web-serving cluster Linux cluster
• DBMS MySQL
• Benchmarks TPC Benchmark

9
Red Hat Cluster Suite

• A cluster of systems to provide highly available
  web services, database servers, and other types
  of services
• Virtual web server configuration
• Using Linux virtual server (LVS) technology

10
A Basic LVS Configuration
11
MySQL Database Services

• Serve highly available data to applications.
• MySQL server packages are installed on each
  cluster system that will run the service.
• Database data is accessed by all cluster members.

12
TPC Benchmark

• TPC defines transaction performance and database
  benchmarks.
• TPC benchmarks measure transaction processing
  and database performance in terms of how many
  transactions a given system and a database can
  perform per unit time.

13
TPC Benchmarks

• TPC-C -- OLTP
• TPC-H decision support for AD hoc queries
• TPC-W web e-commerce
• TPC-R decision support for business reporting

14
References

• IBM Research Report the State of the Art in
  Locally Distributed Web-Server Systems,
• Valeria Cardellini, Emiliano Casalicchio,
  Michele Colajanni
• Www.redhat.Com/manuals/enterprise/RHEL-3-manual/cl
  uster-suite
• Comparing the Memory System Performance of DSS
  Workloads on the HP V-class and SGI Origin 2000,
• Rong Yu, Laxmi Bhuyan, Ravi Iyer
• Building Clustered Linux Systems,
• Robert W. Lucke

15

• A Synthetic Streaming Workload Generator
  Evaluation of Different Streaming Techniques
• Reference GISMO A Generator of Internet
  Streaming
• Media Objects and workloads
• Shudong Jin and Azer
  Bestavros
• Boston University
• A preliminary version appears in ACM SIGMETRICS
  Performance Evaluation Review, November, 2001

16
Workload Charateristics

• Session
• the service initiated by a user's request for a
  transfer and terminated by a user's abortion of
  an ongoing transfer.
• Workload
• Session arrivals
• Properties of individual sessions

17
Session arrival

• Object Populariy
• zipf distribution
• a tendency for requests to be concentrated on a
  few popular objects
• Reference locality
• Heavy-tailed Pareto distribution
• Temporal proximity of requests to the same objects

18
Individual Session

• Object Size
• Lognormal distribution
• User interactivity
• Pareto distribution
• Object encoding characteristics
• Model the VBR auto-correlation of a streaming
  object using a self-similar process
• Use a heavy-tailed marginal distribution to
  specify the level of burstiness of the bit rate

19
Streaming Architecture
20
Base server bandwidth requirments