Server Scalability

1

• Session 11
• Server Scalability Planning for Growth

2
Content Management Issues.

• Naming Policy.
• File and directory naming needs design!
• Names is difficult to 'fix' at a later stage.
• Poor design will cause maintenance grief.
• Content Update Policy.
• Timing who decides.
• Frequency planned?
• Impact on scalability.

3
Content Update Policy

• Without control, a large web system will quickly
  spawn
• Inconsistencies
• Errors
• Inaccessible data etc.
• Do not to up-date on demand!
• Stick to a regular up-date schedule.
• Consider a content management tool. e.g.
  Vignette's StoryServer, see notes.

4
3 Stage Update
5
Scalability Config Issues.

• Scope for initial Webserver Tuning
• Optimise the number of child processes (or
  threads on Windows).
• Review buffer sizes.
• Directives with performance implications
• Use of KeepAlive .
• HostNameLookups on/off/double

6
Scalability Config Issues (2)

• Directives (cont)
• UseCanonicalName on/off/dns
• FollowSymLinks - this is an Option, but can
  cause Apache to waste time checking through the
  file structure.
• Logging of all kinds slows Apache down.
• .htaccess files add overheads.
• Large configuration files also slow Apache, so
  thinning here is a good idea.

7
Scalability Config Issues (3)

• CGI programs influence the performance of the
  website
• Consider FastCGI or mod_perl to speed matters.
• Writing efficient code is always important.
• Other Tricks
• Force popular files to be memory resident.
• Force secure transfers to have more bandwidth.

8
Proxy Servers Performance Issues.

• An Apache Proxy can
• Cache for speed
• Filter for security or decency.
• Apaches proxy functionality is encapsulated in
  mod_proxy
• In order to use mod_proxy, use the ProxyRequests
  directive

to enable - ProxyRequests on or to
disable ProxyRequests off
9
Proxy Customisation.

• To block particular sites from your clients, e.g.

ProxyBlock www.badsite.com
baddomain.co.uk badword

• This will block the specific URL www.badsite.com,
  or the domain, or any URL with names that contain
  badword.

10
Hiding Servers with a Proxy

• Suppose there are two extra servers, parallel to
  the dis server.
• Add the ProxyPass directive to the main
  www.dis.port.ac.uk server configuration file
• so that users.dis.port.ac.uk and
  secure.dis.port.ac.uk appears as directories on
  the main server, e.g.

ProxyPass /users/ http//users.dis.port.ac.uk/ Pro
xyPass /secure/ http//secure.dis.port.ac.uk/
11
Fault Tolerance Clustering.

• Tuning done but insufficient?
• Two basic routes to boosting performance
• Replace the server hardware with more powerful
  boxes.
• Add more servers and distribute the load of
  client requests amongst them.

12
Benefits of multiple servers

• Server machines can be cheaper and easily
  replaceable.
• Individual servers can fall over without the
  website becoming unavailable.
• Increase capacity by adding another server
  synchronising the data.
• No need to alter or reconfigure any of the
  existing servers.

13
Clustering (1).

• Cannot just add an extra servers, would need
  different IP addresses why?
• Set of servers needs to be established as a
  cluster so that
• For external clients it should appear as one big
  fast server with one domain name.
• Clients should not be aware that the load is
  being shared by a cluster of servers.

14
Clustering (2).

• Two basic ways of approaching clustering
• DNS load sharing.
• Web server clustering.
• Several ways for each.

15
DNS Load Sharing.

• Mechanism for synchronising the content on the
  multiple servers!
• Most common approach is Round-Robin DNS
  distribution.
• It works by specifying multiple IP addresses for
  the same host name, e.g. (using a BIND syntax)

www.dis.port.ac.uk. 60 IN A 148.197.203.1 www.dis.
port.ac.uk. 60 IN A 148.197.203.2 www.dis.port.ac.
uk. 60 IN A 148.197.203.3
16
Round Robin DNS Sharing.

• Each DNS request for www.dis.port.ac.uk returns
  the next IP in sequence.
• What about name caches?
• Set a short time-to-live (TTL) the 60
• A lower TTL would
• Improve webserver load sharing,
• But increase the load on DNS server.
• Attraction of the round-robin DNS is its
  simplicity.

17
Round Robin DNS Sharing.

• Not true load balancing, only load sharing.
• The round-robin takes no account of
• Which servers are loaded,
• Which are free,
• Or even which are actually up and running!
• Round-robin DNS makes keeping state for a user
  more difficult eCommerce?

18
Hardware Load Balancing.

• LocalDirector and DistributedDirector were
  products from Cisco (http//www.cisco.com).
• These will rewrite IP headers to redirect a
  connection to a local server.
• See notes for details.

19
Clustering with Apache.

• Last topic on unit!
• Apache provides way to cluster servers using the
  features of mod_rewrite and mod_proxy together.
• This avoids the DNS caching problems and the cost
  of hardware solutions.
• Need a machine as a proxy server, handling
  requests to several back-end servers on which the
  website is actually loaded.

20
Clustering with Apache (2).

• E.g. the proxy takes the name www.dis.port.ac.uk
  and the backend servers might be www1 to www6.
• Wainwright (1999) sets out a method of setting up
  Apache using two parts
• Use mod_rewrite to randomly select a back-end
  server for the client request.
• Use mod_proxys ProxyPassReverse directive to
  disguise the URL of the back-end server.

21
Clustering with Apache (3).

• See notes for details not a practical
  proposition for the unit.
• Awareness of what the proxy server achieves is
  the important point.

22
Summary.

• Content Management Issues.
• Managed Updates for the webserver.
• Configuration issues for scalability.
• Proxy Servers filter and cache!
• DNS (round robin) clustering.
• Hardware clustering.
• Proxy based clustering.

23
Revision.

• Get a suitable printout of your config file
  small font or two columns?
• This years exam is two hours. Plan how you
  intend to use this time!
• Use your config printout to provide config
  examples!