Transcript Document 7468216

Using Global Server Load
Balancing to Improve Web
Site Performance
Jeff Poe, NWS Office of Chief Information Officer
Bruce Marshak, NWS Southern Region Headquarters
NOAATech 2006
November 3, 2005
GSLB in NWS Web Farms
Agenda
•
•
•
•
Problem Overview
Global Server Load Balancing (GSLB)
Summary
Future Plans
NOAATech 2006
GSLB in NWS Web Farms
Problem Overview
• NWS currently has six web farms providing Web
Services (OCIO, NOAA, SRH, ERH, CRH, WRH)
• Different circuit capacities from OC-12 to T3
• Public’s response to certain severe weather events
could use all available circuit capacity at a web farm
– Hurricane landfall for SR
– Severe thunderstorm activity in CR or ER
NOAATech 2006
GSLB in NWS Web Farms
National Doppler Radar Sites
NOAATech 2006
GSLB in NWS Web Farms
Maximum Bandwidth per day in Mbits/sec for 2004
Ivan
Francis
Gaston
Charley
NOAATech 2006
Jeanne
GSLB in NWS Web Farms
Problem Summary
• Web farm bandwidth requirement had doubled each
year for the past 3 years
• Web farms did not have capacity to serve projected
increase for 2005
Strategies to handle increased usage
• Bigger Faster Circuits
• Content Delivery Network a.k.a. Web Caching Service
– Akamai
– Mirror Image Internet
• Global Server Load Balancing
NOAATech 2006
GSLB in NWS Web Farms
GSLB Overview
• GSLB acts as a DNS proxy for selected domain names
• Local server load balancer (SLB) serving requests to web
servers at the web farm for a virtual IP address
• Multiple virtual IP addresses and multiple domain names
• Uses algorithm consisting of several configurable policies and
metrics in order to select best web farm for a user
• Automatically redirects users away from failed sites
• Conceptually aggregates servers and circuit capacity allowing
more efficient use of resources
• Need content mirrored at globally load balanced server farms
NOAATech 2006
GSLB in NWS Web Farms
Domain Name Service (DNS) Operation
• User requests URL, browser queries local DNS to resolve
hostname to IP address
• Local DNS performs queries iteratively, ultimately reaching the
authoritative DNS for requested domain name
• Authoritative DNS replies with one or more IP addresses for a
given domain name
• Local DNS responds back to user’s browser, which opens TCP
connection to given IP address and downloads web pages
• Local DNS caches authoritative DNS response and provides
same address for future requests to domain until time-to-live
(TTL) parameter of domain’s IP specified by authoritative DNS
expires
NOAATech 2006
GSLB in NWS Web Farms
Drawbacks of DNS
• Web server farm supporting website is down or at
capacity limit, authoritative DNS continues to return
that web farms IP address in response to local DNS
queries until authoritative DNS is manually updated
• If application that serves that website is down, users
receive error code, timeout, or stale data
NOAATech 2006
GSLB in NWS Web Farms
DNS with GSLB
• User requests URL, browser queries local DNS to resolve
hostname to IP address
• Local DNS performs queries iteratively, ultimately reaching the
GSLB as the authoritative DNS for requested domain name
• GSLB uses algorithm to select best Virtual IP (VIP) address for a
given domain name and responds to local DNS
• Local DNS responds back to user’s browser, which opens TCP
connection to given VIP address and downloads web pages
• Local DNS caches authoritative DNS response and provides
same address for future requests to domain until time-to-live
(TTL) parameter of domain’s IP specified by authoritative DNS
expires
NOAATech 2006
GSLB in NWS Web Farms
GSLB Policy / Metric processing order
•
•
•
•
Server health check
Weight of VIP
Remote SI's session capacity threshold
Round trip time between remote SI and client
• Geographic location
• Remote SI's available session capacity
• Remote SI's preference value
• Least response selection
NOAATech 2006
GSLB in NWS Web Farms
Milestones
•
•
•
•
•
•
•
•
•
Feb 05: Foundry Training
Mar 05: Installation and Configuration
Apr 05 – May 05: Testing and Proof of Concept
May 05: Serving live data
May 05 – Jun 05: Phased Implementation into Production
Jul 05: Dennis, Emily
Aug 05: Irene, Katrina
Sep 05: Ophelia, Rita
Oct 05: Stan, Wilma
NOAATech 2006
GSLB in NWS Web Farms
Summary
• > 100% increase in peak traffic with ample available
headroom in bandwidth and server capacity
• Average response times < 1 sec
• Improved availability, reliability, and provided a better
service to the public
• Able to take a web farm out of production for
maintenance without impacting users
NOAATech 2006
GSLB in NWS Web Farms
Future Plans
• Additional domains and services
• NOAA web farms
• Norman and Boulder
NOAATech 2006
GSLB in NWS Web Farms
Project Team
•
•
•
•
•
•
•
•
•
•
•
•
•
Susan Beckwith – SRH
Bill Gery – CRH
Alex Hsia – NOAA Boulder
Chi Kang – NOAA Silver Spring
Paul Kirkwood – SRH
Bruce Marshak – SRH
Todd Metcalf – ERH
Reginald Pace – NOAA Silver Spring
Sameka Prather – NWS OCIO
Graham Stork - WRH
Joe Rizo – Foundry
Fortune Vilcko – ERH
Eric Williams - NCIRT
NOAATech 2006
GSLB in NWS Web Farms
Thank You!
NOAATech 2006