Warehouse-Scale Computers and their programming model and workloads. Architectures and cloud computing

1. Warehouse-Scale
Computers
CS4342 Advanced Computer Architecture
Dilum Bandara
Dilum.Bandara@uom.lk
Slides adapted from “Computer Architecture, A Quantitative Approach” by John L.
Hennessy and David A. Patterson, 5th Edition, 2012, MK Publishers and
The Datacenter as a Computer:An Introduction to the Design of Warehouse-Scale
Machines by Luiz André Barroso & Urs Hölzle

2. Outline
 Programming model & workloads
 Architectures
 Cloud computing
2

3. Warehouse-Scale Computers (WSC)
3
www.laserfocusworld.com/articles/print/volume-48/issue-
12/features/optical-technologies-scale-the-datacenter.html http://www.slashgear.com/google-data-center-hd-photos-
hit-where-the-internet-lives-gallery-17252451/

4. WSC (Cont.)
4

5. WSC Layout
5
Source: http://bnrg.cs.berkeley.edu/~randy/Courses/CS294.F07/

6. Main Components of a WSC
6

7. Warehouse-Scale Computer (WSC)
 Provides Internet services
 Search, social networking, online maps, video sharing,
online shopping, email, cloud computing, etc.
 Differences with HPC clusters
 Clusters use higher performance processors & network
 Clusters emphasize thread-level parallelism, WSCs
emphasize request/task-level parallelism
 Differences with datacenters
 Datacenters consolidate different machines & software
into a single location
 Datacenters emphasize virtual machines & hardware
heterogeneity to serve varied customers 7

8. Design Factors for WSC
 Cost-performance
 Small savings add up
 Energy efficiency
 Affects power distribution & cooling
 Work per joule
 Operational costs count
 Power consumption is a primary constraint when
designing a system
 Dependability via redundancy
 Many low-cost components
8

9. Design Factors (Cont.)
 Network I/O
 Interactive & batch processing workloads
 Web search – interactive
 Web indexing – batch
 Ample computational parallelism isn’t important
 Most jobs are totally independent, “Request-level
parallelism”
 Scale – Its opportunities & problems
 Can afford to build customized systems as WSC
require volume purchase
 Frequent failures
9

10. Failure Example
 Consider a WSC with 50,000 nodes. MTTF of a node is 5
years. How many failures be there for a day?
MTTF in days = 5 x 365 = 1,825
Failure rate = 1/1,825 per day
No of failures per day = 50,000/1,825 = 27.4
 Consider a WSC with 50,000 nodes & each node with 4
hard disks. Suppose a annual failure rate of a disk is 4%.
What is the time for a disk failure?
No of disks = 50,000 x 4 = 200,000
No of failures per year = 200,000 x 0.04 = 8,000
Time for failure = 365 x 24 / 8,000 = 1.095 hours/failure 10

11. Programming Models & Workloads
 Batch processing framework
– MapReduce
 Map
 Applies a programmer-
supplied function to each
logical input record
 Runs on thousands of
computers
 Provides new set of (key,
value) pairs as intermediate
values
 Reduce
 Collapses values using
another function 11
http://www.cbsolution.net/techniques/ontarget/mapredu
ce_vs_data_warehouse

12. MapReduce Execution
12
Source: Dean et. al.,
“MapReduce, OSDI, 2004

13. Programming Models & Workloads
(Cont.)
13
www.datanami.com/datanami/2012-07-
16/top_5_challenges_for_hadoop_mapreduce
_in_the_enterprise.html

14. Programming Models & Workloads
(Cont.)
 MapReduce runtime environment schedules
map & reduce task to WSC nodes
 Availability
 Use replicas of data across different servers
 Use relaxed consistency
 No need for all replicas to always agree
 Workload demands
 Often vary considerably
14

15. Computer Architecture of WSC
 Often uses a hierarchy of networks for
interconnection
 Each 19” rack holds 48 1U servers connected to
a rack switch
 Rack switches are uplinked to a switch(es)
higher in hierarchy
 Uplink has 48/n times lower bandwidth –
Oversubscription
 n – No of uplink ports
 Goal is to maximize locality of communication relative
to the rack
15

16. Hierarchy of Switches
16

17. Network Hierarchy
17
Source: www.laserfocusworld.com/articles/print/volume-48/issue-12/features/optical-
technologies-scale-the-datacenter.html

18. Storage Hierarchy
18

19. Infrastructure & Costs
 Location
 Proximity to Internet backbones, electricity cost, property tax rates,
low risk from earthquakes, floods, & hurricanes
 Power distribution
19

20. Power Usage
20
U.S. EPA Report 2007 – 1.5% of total U.S.
power consumption used by data centers
which has more than doubled since 2000 &
costs $4.5 billion

21. How Many Nodes can a WSC Support?
 Each node
 “Nameplate power rating” gives maximum power
consumption
 To get actual, measure power under actual workloads
 Oversubscribe cumulative nodes power by 40%,
but monitor power closely
21

22. Cooling
22
Typically operate around 18 – 22 0C

23. Cooling (Cont.)
23
Cooling system also uses water (evaporation & spills)
e.g. 70,000 to 200,000 gallons per day for an 8 MW facility

24. Efficiency
 Power Utilization Effectiveness (PUE)
= Total facility power / IT equipment power
 ≥ 1
 Median PUE on 2006 study was 1.69
24
Source: http://hightech.lbl.gov/benchmarking-guides/data-a1.html

25. Performance
 Latency is important metric because it is seen by
users
 Bing study
 Users will use search less as response time
increases
 Service Level Objectives (SLOs) & Service Level
Agreements (SLAs)
 Typically given at application level
 e.g., 99% of requests be below 100 ms
 In clouds typically given only for static resources
 CPU speed, no of cores, & memory
25

26. Cost
 Capital expenditures (CAPEX)
 Cost to build a WSC
 Hardware cost dominates
 Operational expenditures (OPEX)
 Cost to operate a WSC
 Power for nodes & cooling dominates
26

27. Cloud Computing
27
Clients
Other
Cloud Services
Govt.
Cloud Services
Private
Cloud
Cloud
Manager
Public Cloud
Green Cloud Computing by Dr. Rajkumar Buyya

28. Cloud Computing (Cont.)
 WSCs offer economies of scale that can’t be
achieved with a datacenter
 5.7 times reduction in storage costs
 7.1 times reduction in administrative costs
 7.3 times reduction in networking costs
 This has given rise to cloud services such as Amazon
Web Services
 “Utility Computing”
 Based on using open source virtual machine & operating
system software
28

29. Amazon Web Services
 Virtual machines
 XEN
 Very low cost
 $ 0.10 per hour per instance
 Primary rely on open source software
 No (initial) service guarantees
 No contract required
 Amazon S3
 Simple Storage Service
 Amazon EC2
 Elastic Computer Cloud 29

30. Amazon Web Services – Example
30
http://www.ryhug.com/free-art-available-on-amazon-amazon-web-services-that-is/