Load Management in Distributed Video Servers

1
Load Management in Distributed Video Servers

• Chaitanya Chemudugunta
• chandra_at_ics.uci.edu

2
Load Management in Distributed Video Servers
By Nalini Venkatasubramanian Srinivas Ramanadhan
International Conference on Distributed Computing
Systems (ICDCS 97), May 1997
3
Overview

• Architecture
• Load Management Mechanisms
• Characterizing Server Resource Usage
• Adaptive Scheduling of Video Objects
• Predictive Placement of Video Objects
• Optimization Methods
• Performance Evaluation

4
A Scalable Video Server Architecture
Distribution Network
requests
data
Distribution Controller
Data Source
Data Source
Data Source
Tertiary Storage
...
control
5
Resources in a Video Server
Client
Client
Network
Processing Module
Communication Modules
Data Manipulation Modules
Storage Modules
6
Load Management Mechanisms

• Replication
• When existing copies cannot serve a new request
• Request Migration
• Unsuitable for distributed video servers
  explicit tear-down and reestablishment of network
  connection.
• Dereplication
• Important Storage space is premium

7
Load Placement Scenario
Data Source S2
Data Source S1
Storage 8 objects Bandwidth 3 requests
Storage 2 objects Bandwidth 8 requests
Access Network
...
Clients
8
Characterizing Server Resource Usage

• Ability to service a request on a server depends
  on
• resource available
• characteristics of a request
• Load factor(LF) for a request
• represents how far a server is from request
  admission threshold.
• LF (Ri, Sj) max (DBi/DBj , Mi/Mj , CPUi/CPUj ,
  Xi/Xj)
• Ri Request for Video Object Vi, Sj Data
  Source j
• DB Disk Bandwidth, M Memory Buffer, CPU CPU
  cycles,
• X Network Transfer Bandwidth

9
Adaptive Scheduling

• When the distribution controller receives a
  request Ri for a video object Vi
• Consider only data sources that have a copy of
  Vi.
• Consider only data sources that have sufficient
  resources to support Ri.
• Chooser server for which LF (Ri, Sj) is a
  minimum.
• If no such server exists
• Reject request.
• Perform replication-on-demand.
• Perform request migration.

10
Predictive Placement of Video Objects

• Determines when, where and how many replicas of a
  video object.
• Initiated periodically.
• Results in an assignment of replicas to data
  sources.
• Formulated as an optimization problem metric to
  be optimized is the total revenue.

11
Predictive Placement of Video Objects Continued

• Each request Ri is associated with a revenue ri.
• ri is dependent on
• Resource required for Ri
• Characteristics of Vi
• Popularity of Vi
• Greedy approach to solve the optimization problem.

12
The Greedy Cost Placement Matrix
PM(Vi, Sj) is the maximum revenue that can accrue
from allocating Vi to Sj.
Greedy heuristic Map(Vi,Sj) 1 if PM(Vi,Sj)
?a ?b max(PM(Va,Sb))
13
Optimizations

• To minimize the overhead of replication
• Eager replication
• Replication of video object in anticipation
• Performed when server resources are free
• Lazy Dereplication
• Critical nature of storage resources
• Mark reusable resources, reclaim disk space later
• If disk blocks are not overwritten, can be
  reclaimed

14
Life of a video object
15
Performance Evaluation Policies
16
Performance Evaluation - Startup Latencies
17
Performance of the basic configuration
p1
p2
p3
p4
18
Performance Evaluation - Varying Replication BW
19
Performance Evaluation Summary

• P1 entails high startup latency, requires high
  storage and replication bandwidth.
• P2 Unacceptably poor performance.
• P3 Similar performance to P4 in many cases.
• At low transfer bandwidths, P4 outperforms P3.
• P4 Performs well in all cases.

20
Thank You