Loading presentation...

Present Remotely

Send the link below via email or IM

Copy

Present to your audience

Start remote presentation

  • Invited audience members will follow you as you navigate and present
  • People invited to a presentation do not need a Prezi account
  • This link expires 10 minutes after you close the presentation
  • A maximum of 30 users can follow your presentation
  • Learn more about this feature in our knowledge base article

Do you really want to delete this prezi?

Neither you, nor the coeditors you shared it with will be able to recover it again.

DeleteCancel

Make your likes visible on Facebook?

Connect your Facebook account to Prezi and let your likes appear on your timeline.
You can change this under Settings & Account at any time.

No, thanks

Energy- and Cost-Efficiency Analysis of ARM-Based Clusters

Analysis of the feasibility of building data centres from ARM based architecture.
by

Michael Reiach

on 16 June 2013

Comments (0)

Please log in to add your comment.

Report abuse

Transcript of Energy- and Cost-Efficiency Analysis of ARM-Based Clusters

Energy- and Cost-Efficiency Analysis of ARM-Based Clusters
Conclusion
Analyzed data from ARM clusters vs Intel workstation
More energy-efficient in the ARM cluster (between 1.21 - 9.5x in various applications)
Multiple ARM processors are needed to provide similar performance to that of Intel workstation
ARM-based data centres are advantageous in lightweight applications

Future work: using more advanced hardware
Cortex A-15, shared resources
Trajectory
General Purpose
Computer Built
Data Centres
Purpose Built
Data Centres
Smartphone
Processor
Development
ARM-Based
Data Centres?
Experiments
Pandaboard
ARM Cortex-A9 MPCore
2 cores
1 GHz
1 GB DDR2
16 GB SSD card
100 Mbps Ethernet
Ubuntu 10.10
1.9 W
Intel Core2 Q9400
4 cores
2.66 GHz
8 GB DDR2
248 GB HDD
1000 Mbps Ethernet
Ubuntu 10.10
95 W
Intel Workstation
Web Server
Throughput
Web Results
Web Server Throughput
In-Memory Database
Video Transcoding
Video
Transcoding
Transcoding Results
In-Memory
Database
Database Reults
x4
Test Methodology
Configuration
Overall power consumption, not just CPUs

Dstat used to record CPU utilization

Energy-Efficiency (EE) =
Performance
Power
Baseline
Basic UInt64 Operations
Completion Times:
Add/Sub- Intel 4-5x faster
Divide- Intel 14x faster

Idle Power:
Pandaboard - 2.59 W
Intel - 26.85 W
Static
Dynamic
Both set up as servers
Powerful workstation client to avoid bottleneck on client side
6 different files sizes
(1 KB - 100 KB)
PHP5 scripts to create HTML dynamically
Under 3 different workloads
(low, medium, high)
10 requests per connection
Static
Dynamic
EE vs CPU % Util
No substantial advantage for ARM when CPU utilization < 20%

EE for ARM cluster = 1.2 - 1.4 when CPU utilization > 20%
Work
Energy
=
Database stored in memory to exclude difference between HDD and SD card
SQLite
Only using one Pandaboard vs Intel Workstation
6 test cases:
10 000 entries insert
5 000 times full table scan with string comparison
Set up an index of the table
2 000 times update with full table string comparison
5 000 times insert from a result of full table scan
10 000 records delete with full table string comparison
ARM at approx. 60% CPU utilization
Intel at approx. 40% CPU utilization

Usually in-memory databases are used for reading rather than writing and updating
Advantage to ARM
Converting from AVI (1920 x 1080) to FLV (640 x 480)
Original video:
1080p, H.264 standard, 24 frames/sec, 86 MB

File split and merged to allow transcoding in parallel on ARM cluster
8 pieces across 4 Pandaboards
= (2 x 10.75 MB)/Pandboard
Result Summary
In all cases, ARM is more energy-efficient
Completion Time
1
2
3
4
1
Processing Capacity
4 × 86 MB
2 × 10.75 MB
= 16
1 Intel workstaion = 16 Pandaboards
ARM performance roof at 2 video pieces, Intel at 4
Most realistic EE of ARM Cluster/Intel is 1.21
No significant energy savings
Cost: Clusters vs Workstations
Data Centres
Total Cost (C)
= (S)evers + (P)ower + (B)uilding + (O)thers
To simplify, infrastructure assumed same in both
cases and
Cost-Efficiency Ratio (R )
CE
15 MW data centre
Cost-Efficiency Ratio (R )
CE
Cost effective when ARM cluster is less than:
Web server - 1.16x price of Intel Workstation
In-memory database - 1.43x
Video transcoding - 1.12
1 Intel:
1 Pandaboard:
Cost: Individual Processors
p = % of cost for processor (20%, 50%)
U= Unit prices
N= # of ARM CPUs to equal perf. of 1 Intel
Cost-effective for ARM when Intel is (p=0.2):
Web server - 6.64x more expensive
Database - 1.33x
Video transcoding - 9.95x
References
[1] Z. Ou; B. Pang; Y. Deng; J.K. Nurminen; A. Ylä-Jääski; P.Hui; , "Energy- and Cost-Efficiency Analysis of ARM-Based Clusters," Cluster, Cloud and Grid Computing (CCGrid), 2012 12th IEEE/ACM International Symposium on , vol., no., pp.115-123, 13-16 May 2012

[2] J. Hamilton, Cooperative expendable micro-slice servers (CEMS): low cost, low power servers for Internet-scale ser- vices, in CIDR ’09, 8 pages.
[2]
[1]
[1]
[1]
[1]
[1]
[1]
[1]
[1]
[1]
[1]
[1]
In the News
Motivation
Experiments
Results
Feasibility
Conclusion
Full transcript