More Related Content
Similar to The DevOps PaaS Infusion - May meetup
Similar to The DevOps PaaS Infusion - May meetup (20)
The DevOps PaaS Infusion - May meetup
- 1. Gary Berger
Technical Leader, Engineering Office of the CTO
May 17, 2012
© 2010 Cisco and/or its affiliates. All rights reserved. Cisco Confidential 1
- 2. Technical Leader, Office of the CTO Data Center
Business Unit
• 22 Years Infrastructure Architecture and Platform
Development
• Performance and Capacity Planning
• Data Center Design
• Protocol Architecture
• Application Design and Scalability
• Software Defined Networking
@gbatcisco
garyberger.net
© 2010 Cisco and/or its affiliates. All rights reserved. 2
- 3. • Partnering since 2008
• Advanced integration with Cisco
Unified Compute System
• OpenStack Integration (Nova,
Quantum)
• “Cloud in a Box” - High performance
scaling to 1TB and 40 Cores.
© 2010 Cisco and/or its affiliates. All rights reserved. 3
- 4. Data Size compared to Task Rate
1. Compute Intensive
• Low number of tasks and small input size Data Size
• This includes MPI workloads familiar in HPC applications.
High
2. Data Analytics
• Larger data sizes familiar to Map/Reduce programming
model
Analytics
3. Loosely Coupled Data Intensive
Med
• Modest data size but increasing the number of tasks
• Indicative of data-grid applications and HTC which are
bounded by memory capacity but also can be bounded by Compute
Intensive
local disk I/O Loosely Coupled
4. Data Intensive Low
• Many tasks and large datasets.
• Formidable challenge for networks with dense matrix 1 1K 1M
• Categorized as Many Task Computing (MTC) Number of Tasks
© 2010 Cisco and/or its affiliates. All rights reserved. 4
- 5. • Current Internet Trends
• Quick historical perspective and state of the “cloud”
• Data Center as a Business Archetypes
• Mechanical Sympathy
• Real World Challenges
• Service Centric Networking
© 2010 Cisco and/or its affiliates. All rights reserved. 5
- 6. • +900M Users • +150M Active Users • 4B videos view/day
• 3.2B Likes/Comments/day • +340M Tweets per day • 800M visitors/mnth
• +300M photos uploaded/day • 60H uploaded/min
• 125B Friendships
© 2010 Cisco and/or its affiliates. All rights reserved. 6
- 7. Mobile Data Traffic Mobile Data Transfer Distribution
(Exabytes/Month) 100%
12 90%
80%
10
70%
8 60% Other
6 50% Web
40%
4 Video
30%
2 20%
0 10%
2011 2012 2013 2014 2015 2016 0%
Operator A Operator B Operator C Operator D
Source: Cisco VNI Mobile 2012 Source: ByteMobile Mobile Analytics Report 2012
© 2010 Cisco and/or its affiliates. All rights reserved. 7
- 9. Alan Turing
June 1912 - June 1954
© 2010 Cisco and/or its affiliates. All rights reserved. 9
- 10. Host Centric Client Centric Database Centric Web Centric Service Centric
“Technical Debt” “New Economy”
• Time shared • Desktop • Evolution of Client/ • Normalized • Loosely coupled
system applications Server Presentation Layer components
• Explicit control • Centralized File & • 4GL Programming • Ubiquitous Access • Web based
• Restricted scope Print • Stored Procedures • Ubiquitous API interactions
• Tightly Coupled • Many dependencies • Vertically Integrated • Self-Described Data • Almost Infinite
• Vertically • Low network • Proprietary Scalability
Integrated utilization • Global scope
• App driven
Sparse to Dense operational integrity
© 2010 Cisco and/or its affiliates. All rights reserved. 10
- 11. © 2010 Cisco and/or its affiliates. All rights reserved. 11
- 13. Geographic Market
Expansion
Reach
Your Business
Service
New Sources Monetization
Of Data
Capex
Controls
© 2010 Cisco and/or its affiliates. All rights reserved. 13
- 14. © 2010 Cisco and/or its affiliates. All rights reserved. 14
- 15. “Until now, cloud computing has been mostly about the
distribution of applications”
“The next wave of cloud computing will enable the
sharing of the environment to run those applications.”
“You will be able to take advantage of what we had to
build in order to create those applications”
Ben Fried, CIO Google 2012
© 2010 Cisco and/or its affiliates. All rights reserved. 15
- 16. © 2010 Cisco and/or its affiliates. All rights reserved. 16
- 17. Homogenous Web Scale Heterogeneous Multi-Tenant Unified Multi-Service
• Highly distributed • Highly virtualized • Highly flexible
• Leverages scale-out/parallel • Leverage compute arbitrage and • Incorporates qualities of both HMT and
application design SPOT market HWS
• Minimizes heterogeneous applications • Benefits from a mixture of customer • Purpose built to remove infrastructure
by providing higher level services and market segments to randomize barriers to application development
common resources management demand • Manages resources more efficiently by
• Enhanced focus on cost and efficiency • Complex engineering due to controlling allocation via higher-level
due to large population. overlapping naming/addressing platform services
• Operational separation of code, data, • Complex operations due to • Provides best ROI and flexibility
configuration and policy uncoordinated modifications, through common abstraction libraries
interference due to competing access and runtimes
to shared resources • “Its all about the app”
• Enhanced focus on security and • Operations as a Service
isolation
Examples: Google, MSFT, Facebook, Examples: Amazon EC2, Rackspace, Examples: Amazon (DDB, EMR), RHEL
Yahoo etc..). OpenShift, MSFT Azure, VMForce
© 2010 Cisco and/or its affiliates. All rights reserved. 17
- 18. Having an understanding of the underlying architecture and behavior in order to build
better systems.
Power Wall I/O Wall App Memory Wall
© 2010 Cisco and/or its affiliates. All rights reserved. 18
- 19. Coherency starts to force retrograde behavior
O(N^2)
Serialized Contention
starts to dominate (i.e.
locking)
Amdahl
Linear Growth
p
(Scale-Up/In)
C( p) =
1 + α ( p −1) + β p( p −1)
© 2010 Cisco and/or its affiliates. All rights reserved. 19
- 20. Load
Balancer
Load
Load Web Balancer Firewall
Network Balancer
Network
Network
Network
Firewall Firewall DBA
Presentation App App
Tier Logic Data
Increased Delay/Limited Scalability
© 2010 Cisco and/or its affiliates. All rights reserved. 20
- 21. Cluster Manager
Recipe
Caching
App Data
&
Services Services
SDN Controller Presentation
© 2010 Cisco and/or its affiliates. All rights reserved. 21
- 22. network{
name: publish_subscribe
application {
qos: best_effort
name : myApp
isolation: per_domain
tenantID: tenantID
encryption: true
service {
msgPattern: pubsub
compute {
}
template: ucs_small_linux
storage {
}
name= cache_persistent
network {
cache {
template: publish_subscribe
capacity: 5G
}
evictionPolicy: LRU
storage {
}
template: cache_persistant
persistence{
}
block: 10TB
file: extfs
}
RAID: 10
}
}
}
© 2010 Cisco and/or its affiliates. All rights reserved. 22
- 23. • Effective Resource Sharing
• Further away from the metal, the harder it is to understand (non-deterministic performance)
• Contention grows while accessing shared resources
• What instruments to collect analyze and model
• Programming Languages
• Generally languages are insufficient for building large applications (lack of procedures in JAVA, lack of encapsulation in
Python, etc.)
• Concurrency is still extremely difficult and hard to reason about (trend towards functional reactive programing)
• Throw away code
• Network Scalability
• Segmentation and Isolation
• Address Learning
• Application aware
• Programmatic Interfaces
• Security
• In-flight/At-Rest encryption
• Proper tradeoff between performance and privacy
• Rat-Hole because of lack of tools, developer education and highly incentivized and motivated hacker community
© 2010 Cisco and/or its affiliates. All rights reserved. 23