Analysis of Caching and Replication Strategies for Web Applications

1
Analysis of Caching and Replication Strategies
for Web Applications
Authors Swaminathan Sivasubramaniam, Guillaume
Pierre, Maarten van Steen.

• Presented By
• Sudarsan Maddi
• Graduate Student

2
Topics That We Will be Seeing

• Introduction
• Techniques to scale Web applications
• Performance Analysis
• Choosing the Right Strategy

3
Introduction

• In this paper the authors present qualitative and
  quantitative analysis of replication and caching
  techniques to host Web applications.
• Their analysis shows that selecting the best
  mechanism depends heavily on data workload and
  application characteristics.

4
Introduction

• Web sites are slow dew to many reasons, one of
  the main reason is dynamic generation of web
  documents.
• Web page caching Fragments of HTML pages the
  application generates are cached to serve future
  requests.
• Content-delivery networks such as Akamai do this
  by deploying edge servers around the Internet
  thus reducing requests network latency.

5
Introduction

• Limitations of page caching have given raise to
  different approaches for scalable Web
  applications, classified broadly into
• Application code replication
• Cache database records
• Cache query results
• Entire Database replication
• In this article they have given overview of
  various scalable techniques compared and analyzed
  their features and performance.

6
Techniques to scale Web Applications

• The techniques we are going to see are
• Edge Computing
• Data Replication
• Content-Aware data Caching (CAC)
• Content-Blind data Caching (CBC)

7
Edge Computing

• In this the application code is replicated at
  multiple edge servers and data is centralized.
• Akamai and ACDN use this technique.
• The data centralization create problems,
• If the edge servers are located worldwide, each
  data access incurs WAN latency.
• The central database becomes a performance
  bottleneck if the load increases.

8
Data Replication

• Solution for Edge computing is to place the data
  at each edge server.
• Database replication (REPL) techniques can help
  maintaining identical copies at multiple
  locations.

Continued
9
Data Replication

• The problem with this is when there is a database
  update.
• This creates huge network traffic and performance
  overhead.

10
Content-Aware data Caching (CAC)

• Instead of maintaining full copies of database
  CAC systems cache database query results as the
  application code issues them.
• Query Containment Check The application running
  at the edge-server issues a query, the local
  database checks if it has enough data to answer
  the query locally.
• Containment check results positive query is
  present locally, else its sent to central
  database and inserts the result in its local
  database.

Continued
11
An Example of CAC

• CAC store query results efficiently
• For example
• Query Q1 Select from items where pricelt50
• Query Q2 Select from items where pricelt20
• Query template QT1
• Select from items where pricelt

12
Content-Aware data Caching (CAC)

• This query containment check is highly
  computationally expensive because it must check
  the new query with all previously cached queries.
• In order to reduce this cost CAC makes use of
  query template, which is a parameterized SQL
  query whose parameter values are parse at runtime
• In, CAC systems update queries is always
  executed at the central database.

13
Content-Blind data Caching (CBC)

• Here, edge servers dont need to run a database
  at all.
• Instead they store the results of remote database
  queries independently.
• The query results aren't merged here storing
  redundant information, and will have a hit only
  if application issues exact query, so hit rates
  are low

Continued.
14
Content-Blind data Caching (CBC)

• This have some advantages over CAC as,
• Incurs very little computational load.
• Caching query results as result sets instead of
  database records, so can return results
  immediately.
• Finally, inserting a new element into the cache
  doesn't require a query rewrite.

15
Scalable Web hosting.  (a) edge computing, (b)
content-aware caching, (c) content-blind
caching, and (d) data replication.
16
Performance Analysis

• To compare the four techniques, they have made
  use of two different applications,
• RUBBoS, a bulletin-board benchmark application
  that models Slashdot.org,
  http//jmob.objectweb.org/rubbos.html
• TPC-W, an industry-standard e-commerce benchmark
  that models an online book store such as
  Amazon.com,
• http//pgfoundry.org/projects/tpc-w-php/

17
Performance Analysis

• They have measured the end-to-end client latency,
  which is the sum of network latency and internal
  latency.
• The results shows that CBC performed best in
  terms of client latency whereas EC performed the
  worst for RUBBoS.
• Were as for TPC-W REPL performed the best and EC
  worst again.

18
Performance Results
(a) RUBBoS benchmark (b) TPC-W Browsing (c) TPC-W
Ordering
19
Choosing the Right Strategy

• According to the author the Web designers should
  choose the scalable technique by carefully
  analyzing their Web application characteristics.
• They have suggested the best strategy is the one
  that minimizes the applications end-to-end client
  latency.
• This latency is affected by many parameters as
  hit ratio, database query execution time,
  application server execution time.
• To do this they have proposed a concept called
  virtual caches (VC).

Continued
20
Choosing the Right Strategy

• VC behaves just like a real cache but it stores
  only meta data, such as the list of objects in
  the cache, sizes. So this requires less memory
  compared to real caches.
• So with the help of these VC we can get the hit
  ratios and execution times for servers and can
  estimate end-to-end latency.

21
Thank You.