Distributed Systems and Multimedia

Load Management in
Distributed Video Servers
Chaitanya Chemudugunta
[email protected]
Load Management in
Distributed Video Servers
By
Nalini Venkatasubramanian
Srinivas Ramanadhan
International Conference on Distributed Computing Systems
(ICDCS 97), May 1997
Overview
Architecture
Load Management Mechanisms
Characterizing Server Resource Usage
Adaptive Scheduling of Video Objects
Predictive Placement of Video Objects
Optimization Methods
Performance Evaluation
Load Management in Distributed
Video Servers
3
A Scalable Video Server
Architecture
Distribution Network
requests
Distribution
Controller
data
Data
Source
Data
Source
...
Data
Source
Tertiary
Storage
control
Load Management in Distributed
Video Servers
4
Resources in a Video Server
Client
Client
Network
Communication
Modules
Data Manipulation
Modules
Processing
Module
Storage
Modules
Load Management in Distributed
Video Servers
5
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
Load Management in Distributed
Video Servers
6
Load Placement Scenario
Data
Source
S2
Data
Source
S1
Storage:
8 objects
Bandwidth:
3 requests
Storage:
2 objects
Bandwidth:
8 requests
Access Network
...
Clients
Load Management in Distributed
Video Servers
7
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
Load Management in Distributed
Video Servers
8
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.
Load Management in Distributed
Video Servers
9
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.
Load Management in Distributed
Video Servers
10
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.
Load Management in Distributed
Video Servers
11
The Greedy Cost Placement
Matrix
S1
S2
V1
min(N1,1/LF(R1,S1))*r1
min(N1,1/LF(R1,S2))*r1
V2
min(N2,1/LF(R2,S1))*r2
min(N2,1/LF(R2,S2))*r2
V3
min(N3,1/LF(R3,S1))*r3
min(N3,1/LF(R3,S3))*r3
max(PM(Vi,S1))
max(PM(Vi,S2))
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))
Load Management in Distributed
Video Servers
12
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
Load Management in Distributed
Video Servers
13
Life of a video object
Load Management in Distributed
Video Servers
14
Performance Evaluation
Policies
Policy
Placement of Video Objects
Request Scheduling
P1
Replication on-demand
Adaptive
P2
Predictive placement
Predictive
P3
Predictive placement
Adaptive
P4
Predictive placement + eager
replication
Adaptive
Load Management in Distributed
Video Servers
15
Performance Evaluation Startup Latencies
Load Management in Distributed
Video Servers
16
Performance of the basic
configuration
p1
p2
p3
p4
Load Management in Distributed
Video Servers
17
Performance Evaluation Varying Replication BW
Load Management in Distributed
Video Servers
18
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.
Load Management in Distributed
Video Servers
19
Thank You