Upload
akamaidevrel
View
309
Download
1
Embed Size (px)
Citation preview
© AKAMAI - EDGE 2016
Solving Everyday Problems with Next Generation Mapping
Liz Borowsky, VP Platform Engineering, Akamai Technologies
© AKAMAI - EDGE 2016
What do I mean by mapping anyway?
© AKAMAI - EDGE 2016
A small scale example
One Datacenter
© AKAMAI - EDGE 2016
Mapping: Load Balancing on steroids
© AKAMAI - EDGE 2016
Mapping Fundamentals
Availability
• Content is always accessible• Even when machines fail• Even when routing changes• Even when individual networks have
problems
Performance
• The best Quality of Experience• Low latency• High throughput • Plenty of machine resources
Primary Problem a CDN Mapping system wants to solve:
© AKAMAI - EDGE 2016
The two main approaches to mapping
AnycastedNameserver Mapped
4.3.2.1
© AKAMAI - EDGE 2016
Anycast: Features
• Conceptually simple • Easy to scale
(to a point)
Inherently somewhat centralized
© AKAMAI - EDGE 2016
Anycast: Availability
BGP convergence time 30 seconds to up to 5 minutes*
* https://labs.ripe.net/Members/vastur/the-shape-of-a-bgp-update
© AKAMAI - EDGE 2016
Anycast: Performance. US data, Gomez Last Mile tests
© AKAMAI - EDGE 2016
Anycast: Challenges – Lack of control
• Load balancing done by BGP tuning• Hard to use resources efficiently• At the mercy of BGP churn and convergence• Scales only to a point
© AKAMAI - EDGE 2016
Nameserver Mapping: Availability
Better Control20 seconds
to 1 min
© AKAMAI - EDGE 2016
Nameserver Mapping: Performance
Control enables the abilityto be widely
distributed
© AKAMAI - EDGE 2016
Challenge: Nameservers far from end users
Non-optimal mappingPoor performance
© AKAMAI - EDGE 2016
Great performance
Solution: ECS (EDNS0-Client-Subnet DNS Extension*)
*https://trac.tools.ietf.org/html/rfc7871
© AKAMAI - EDGE 2016
Solution: ECS (EDNS0-Client-Subnet DNS Extension)
Average improvements from enablement with openDNS and GoogleDNS
© AKAMAI - EDGE 2016
Challenge: Scaling the internet
Control = Opportunity
© AKAMAI - EDGE 2016
Challenge: Mobile End Users
End users go throughmobile gateways
Combine clustering with ECS, but use the Gateway IP Subnet
© AKAMAI - EDGE 2016
Start of a solution: Mobile gateway clustering
© AKAMAI - EDGE 2016
Challenge: Scaling the internet
Control = Opportunity
© AKAMAI - EDGE 2016
Challenge: Congested peering on the global internet
Distribute loadAvoid congestion
© AKAMAI - EDGE 2016
Distribute load away from congested peering
• Measure throughput• Map away if throughput degrades• Use distributed platform to get content close to the end user• Offload overloaded peering links
© AKAMAI - EDGE 2016
Early results
© AKAMAI - EDGE 2016
Challenge: Congested peering on the global internet
Distribute loadAvoid congestion
© AKAMAI - EDGE 2016
The Future
With control comes opportunity…• ECS in more networks• More metrics and data• More control to give better performance• More control to help the internet scale• Strategic anycast – best of both worlds
© AKAMAI - EDGE 2016
Questions?
問題Fragen
שאלותdomande
質問
질문
frågor
preguntas
Вопросыpytania
األسئلة
vrae
Sorular
spørsmål