High Performance Web Server

1
High Performance Web Server

• NTUIM
• R89725018 Chen Pei-wen
• R89725013 Cheng Pei-chun

2
Outline

• Introduction
• Load balancing
• Content-based Switching
• Implementation Architecture
• Conclusion
• Reference

3
Introduction

• Performance and high availability are critical at
  web sites which receive large number of requests.
• The QoS a web server provides to end users
  depends on
• Network-transfer speed
• Server-response time

4
Introduction(cond)

• Network-transfer speed is mainly a matter of
  Internet-link bandwidth.
• Server-response time depends upon available
  resources
• Single server
• Multiple servers

5
Introduction(cond)

• Single server

Web-server Software
Well-designed server process Adjust web-server
software
Operation System
Specific operation system Adjust operation-system
parameter
Hardware
Install more RAM Replace the CPU with a faster
one Use faster SCSI controllers and disks
6
Introduction(cond)

• Multiple Servers
• Improve performance by increasing the number of
  Web servers.
• This involves an attempt to distribute the
  traffic onto a cluster of back-end Web servers.
• Load balancing is needed.

7
Load Balancing

• Goal
• To balance the traffic onto available server
• The technical distribution is totally transparent
  to the end user

8
Load Balancing(cond)

• Benefits
• Improve reliability (fault tolerance)
• If you are using a single server and it fails,
  the site goes down with it.
• This is especially bad for e-commerce and
  financial sites which lose money if they are out
  of service.
• With a load balanced group of servers, loss of a
  single server will only slightly affect overall
  site performance and the site will not go down.

9
Load Balancing(cond)

• Improve performance
• Load balancing allows multiple servers to be
  available to handle larger number of incoming
  client requests.
• Lower cost
• With load balancing providing fault tolerance to
  the entire site, the reliability of each
  individual server is less critical.
• We can use lower-cost servers without
  compromising overall reliability.

10
Load Balancing(cond)

• Improve scalability and flexibility
• With one server, all you can do if your traffic
  increases is upgrade that server or buy a bigger
  one.
• With load balanced groups of servers, you can
  simply add more servers gracefully to server
  farm.
• Improve maintainability
• The flexibility of a load balanced group of
  servers allows you to remove individual servers
  from service for repair or upgrade without
  affecting the overall availability of the site.

11
Load Balancing(cond)

• Load balancing algorithm (RFC 2391)
• Round-Robin
• This is the simplest scheme, where a host is
  selected simply on a round robin basis, without
  regard to load on the host.
• Least Load first (session count)
• the host with least number of sessions bound to
  it is selected to service a new session.
• Each session is assumed to be as resource
  consuming as any other session

12
Load Balancing(cond)

• 3. Least traffic first (bytes or packet count)
• measure system load by tracking packet count or
  byte count directed from or to each of the member
  hosts over a period of time.
• 4. Least weighted load
• Weights to sessions, based on likely resource
  consumption estimates of session types
• Weights to hosts based on resource availability.
• 5. Fastest response
• periodically ping member hosts and measure the
  response time to determine how busy the hosts
  really are

13
Load Balancing - RR DNS

• Round-Robin DNS Approach
• Allows a single domain name to be associated with
  several IP addresses
• Example using CNAME (canonical name) resource
  records
• www.foo.dom. IN CNAME www1.foo.dom
• IN CNAME www2.foo.dom
• IN CNAME www3.foo.dom
• IN CNAME www4.foo.dom

14
Load Balancing - RR DNS

• Operation
• A browser has to take to retrieve the URL is to
  resolve the corresponding IP address
• A name resolver that calls a nearby DNS server,
  which then actively iterates over the distributed
  DNS server hierarchy on the Internet until it
  reaches the Round-Robin DNS server, which finally
  gives the IP address
• A browser takes the IP address and create a
  connection with the assigned server

15
Load Balancing - RR DNS
16
Load Balancing - RR DNS

• Attractiveness of Round-Robin DNS
• The concept is simple
• It requires no additional hardware
• Drawbacks of Round-Robin DNS
• DNS is unaware of the status of web servers
• All servers are assumed to have equal capability
  to offer all services

17
Load Balancing - RR DNS

• The caching of DNS data can cause load imbalances
• In practice, DNS servers cache the resolved data
  at any point in the DNS hierarchy both to
  decrease the resolver traffic and to speed up
  resolving.

Database
Database
User process
query
query
query
Name resolver
Name Server
RR DNS Server
response
response
response
cache
cache
cache
18
Load Balancing - L4 Switch

• Layer-4 Switch Approach
• These switches sit between the connection to the
  Internet and the server farm

19
Load Balancing - L4 Switch
20
Load Balancing - L4 Switch

• Operation
• The switch recognizes when a client is requesting
  a new session by identifying the TCP SYN packet
• The request is forwarded to the best available
  server based on the configured load balancing
  algorithm
• The switch maintains a session-server binding
  table that associates each active session with
  the real server to which it is assigned

21
Load Balancing - L4 Switch

• It performs address substitution so that the real
  server will transparently receive packets for
  that session
• Likewise, the switch intercepts packets traveling
  from the real server to the client and performs
  the reverse address substitution
• The switch recognizes when the session is
  terminated by identifying the TCP FIN packet
• Then it removes the session-server binding from
  its binding table

22
Load Balancing - L4 Switch

• Attractiveness of Layer-4 Switch
• Good load balancing can be achieved
• No problem of the caching of DNS data
• Sophisticated algorithm can be used
• Aware of the failures of web servers
• Limitation of Layer-4 Switch
• It has no concept of what content is being
  requested

23
Load Balancing - L4 Switch

• All content should be replicated
• Cache hit rate may be low

24
Load Balancing

• Benefits of Content Awareness

25
Content-based Switching

• Content-based Switching
• Intelligently load balances traffic across
  delivery nodes, dynamically directing specific
  content requests to the best site and server at
  that moment.
• Based on content availability, application
  availability and server load.
• Adds protection against flash crowds and ensures
  transaction continuity for e-commerce
  applications.
• Enables advanced personalization and
  prioritization for important content and
  customers.

26
Content-based Switching(cond)

• Benefits
• Increased performance due to improved hit rates
  in the back-ends main memory caches.
• Increased secondary storage scalability due to
  the ability to partition the servers database
  over the different back-end nodes
• The ability to back-end nodes that are
  specialized for certain types of requests

27
Content-based Switching(cond)

• Products
• ArrowPoint's Content Smart Web SwitchesWebNS
• Foundry networksServerIron Traffic Management
  system Internet IronWare
• Nortel NetworkssAccelar Load Balancing Server
  Switch

28
Content-based Switching(cond)

• Switch Architecture

29
Content-based Switching(cond)

• Idea

User makes a content request by typing a URL
into a Browser.
Web switch with virtual IP of the requested URL
intercepts the request.
Web switch spoofs TCP connect back to client and
examines packet URLs.
Simultaneously, the Web switch examines HTTP
header and URL and compares to current content
rules to select best server or cache to satisfy
request.
A flow is created between the switch and the
optimal server and "snaps" together with the flow
from the client to the switch.
A flow control block is created in the port ASIC
and all subsequent packets are forwarded without
intervention by the switch controllers.
30
Content-based Switching(cond)
31
Content-based Switching(cond)
32
Content-based Switching(cond)
33
Content-based Switching(cond)

• Variant Switching Scheme
• URL Switching Directs HTTP requests to a group
  of servers using information contained in URL
  string.
• Greater control over the website deployment to
  place different web content on different servers
• Eliminating unnecessary duplication of all
  content across all load-balanced servers.
• Ex Different file types Different request

34
Content-based Switching(cond)

• Variant Switching Scheme
• Cookie SwitchingDirects HTTP requests to a
  server group based on information embedded in a
  cookie in the HTTP header.
• Cookie specifies which server group should handle
  the request.
• Ensures that a particular server group always
  handles requests from a particular client even
  across sessions.
• Guarantee persistent end-user experience.
• Ex Personalized web page Prioritized service

35
Content-based Switching(cond)

• Variant Switching Scheme
• SSL Session ID Switching All the SSL connections
  between a client and server must reach the same
  host.
• Ensures that all the traffic for a SSL
  transaction with a given SSL-ID always goes to
  the same server.
• Key feature for commerce, financial web sites
• Ex Prevent shopping cart loss Access
  control Prevent source address overload

36
Content-based Switching(cond)

• Load balancer evaluation criteria
• Plans for Layer 3 or Layer 4 switching
• Number of servers and planned growth
• Type of content to be balanced
• Number of server sites to be balanced
• Sophistication of balancing algorithms
• Degree of fault tolerance required
• Interfaces and port density
• Support requirements

37
Implementation Architecture

• Design, Implementation and Performance of a
• Content-Based Switch
• Infocom 2000
• George Apostolopoulos, David Aubespin, Vinod
  Peris, Prashant Pradhan, Debanjan Saha

38
Implementation Architecture(cond)

• Switch Architecture

• The Layer 5 system consist of a switch core to
  which a number of custom built intelligent port
  controller are attached.
• Layer 5 functions, such as the parsing of HTTP
  protocol messages and URL based routing, are
  performed by the processor.

39
Implementation Architecture(cond)

• Switch Architecture

• Port controller identify the packets that need to
  be handled by the processor and forward them to
  the processor.
• Make sure it can achieve very high speed while
  delivering sophisticated Layer 5 functionality.

40
Implementation Architecture(cond)

• Operation Blueprint

• Phase 1 it intercepts the TCP connection setup
  request from the client and response by
  establishing a connection to the client.
• Phase 2after routing decision is made, it sets
  up a second connection to the appropriate server
  node.
• Phase 3splicing the two TCP connections

41
Implementation Architecture(cond)

• Processing at Port Controllers

42
Implementation Architecture(cond)

• Processing at CPU

• CPU acts as the end-points for the TCP
  connections to the client and the server until.
  they are spliced.
• Splices the connection by sending the appropriate
  control messages to the port controllers.
• Handling of TCP options deserves special
  attention.
• Reject all TCP options
• Enumerate the minimum set of options supported by
  all nodes

43
Implementation Architecture(cond)

• URL look up

• To be able to dispatch HTTP requests based on
  URLs, the L5 system has to know mapping from the
  URL to the web server on which the page resides.
• Use a hash function and set default size of all
  hash buckets to 256.

44
Conclusion

• The concept of content-based switching is
  understandable, but efforts are needed to
  implement it well.
• Content based-service differentiation can be used
  to provide service differentiation based on the
  user profiles.
• Not only load balancing but also persistence pays
  in E-Commerce.

45
Reference

• RFC 2391 LSNAT
• Webtechniques Load balancing your web
  siteshttp//www.webtechniques.com/archives/1998/0
  5/engelschall/
• HydraWEB Load Balancinghttp//www.hydraweb.com/lo
  ad_balancing/index.asp
• Techniques for Designing High-Performance Web
  Siteshttp//www.research.ibm.com/people/i/iyengar
  /ieeeic/ieeeic.html
• Locality-Aware Request Distribution in
  Cluster-based Network Service In Architectural
  Support for Programming Languages and Operating
  System, 1998
• TCP/IP Related Protocols 2 edition Uyless
  BlackChapter four The Domain Name System

46
Reference (cond)

• Foundry Products Application Noteshttp//www.foun
  drynet.com/appnotes.html
• Alteon WebSystems Web Switching White
  Paperhttp//www.alteonwebsystems.com/products/whi
  tepapers/index.asp
• Design, Implementation and Performance of a
  Content-Based Switch Infocom 2000George
  Apostolopoulos, David Aubespin, Vinod Peris,
  Prashant Pradhan, Debanjan Saha
• Cisco CDNshttp//www.cisco.com/warp/public/779/la
  rgeent/learn/technologies/content_networking/

47
URL Switching
Back
48
Cookie Switching
Back
49
SSL Session ID Switching
Back