The document discusses interdisciplinary research in cloud computing, highlighting opportunities for improvements and applications across various areas, including computer architecture, physics, and data processing. It presents specific projects and solutions leveraging cloud technology for tasks like seismic imaging, virial coefficient calculations, and educational content management. It also addresses the implications of new memory technologies and their impact on programming models and data security.

1. Interdisciplinary Research for Cloud
Computing: Future and challenges
Rodolfo Azevedo
rodolfo@ic.unicamp.br
University of Campinas - UNICAMP

2. Background
Porto Alegre, July 6th, 2016 2
Computer
Architecture
Embedded
Systems
Power
Consumption
New Memory
Technologies
Technology for
Education
Simulation:
Molecular &
Seismic
Programming
Models
Social Impact

3. OpenPOWER Foundation
• First Latin America University Member
• Goal: Evaluate and improve an open computing
ecosystem
• Resources: Local OpenStack cloud with 6 nodes
 up to 720 virtual machines
 Joint laboratory at the campus
• Opportunities:
 Coherence Accelerator Processor Interface
(CAPI)
 Fully programmable software stack. Open
source firmware, switches, operating system
 Full Debian Linux (little endian) distribution
ported to POWER architecture
http://openpower.ic.unicamp.br
int main()
{
int b = 2;
int a;
for(a=-5000;a<-2;a=a >> 1)
{
b = b*a;
}
return b;
}
Example Project
Intel x86 POWER
?
=
Porto Alegre, July 6th, 2016 3
SBESC 2015

4. Calculation of Virial Coefficients
Order: n
Process all m
graphs
Find all biconnected
graphs with n vertices
Result: Coefficient Bn
(single number)
• Physics Problem: Understanding macroscopic behavior of fluids from microscopic
interactions.
• Approach: Assessment of convergence properties of virial expansion series.
• Physical system: Uhlenbeck-Ford model for which virial coefficients can be computed exactly.
• Computational approach:
• Problem:
• Number of graphs grows exponentially with n.
• Calculation of B11 (largest known coefficient) required 1 week
of processing on 120 CPU cores.
• Coefficient B12 would require ~170 times longer (> 3 years!).
• Solutions:
• Code optimization: B12 took 9 days in one core on a
notebook
• Better parallelization: B13 took 15 days in ~400 cores
• B14 is 500x bigger  GPU & Cloud
Porto Alegre, July 6th, 2016 4

5. Imaging the earth’s interior
• Many human activities (engineering, natural resources, environment) rely on a correct
understanding of the earth’s interior
• Efficient construction of subsurface images is of paramount importance
• Bag of tasks, small number of data transfers
Seismic
dataset
Processing
& Imaging
Verilog
OpenCL
GPU
FPGA
78 Semblances/Joule
118 Semblances/Joule
How good?
months
days/weeks
Xeon E5-2630 2.6GHz
8 Semblances/Joule
Porto Alegre, July 6th, 2016 5

6. OpenMP to the Cloud
• Using existing programming model: OpenMP
• Move tasks to the cloud for scalability
Porto Alegre, July 6th, 2016 6
#pragma omp parallel for
for (i = 0; i < MAX; i++)
Compute(i);
Compiler
Local
Cloud
SBAC-PAD 2014

7. WebLectures Project
• More than 1500 lectures recorded in video
since 2009
– No support for big videos on YouTube at that time
– Local server, local infrastructure
• New opportunities on the cloud
– Voice recognition
– OCR
– Automatic creation of cross-disciplines references
Porto Alegre, July 6th, 2016 7
WIPTE 2008, 2009

8. DoeNota Project
• São Paulo state reimburses a small part of sales tax
– Goal: increase the number of people that asks for receipts 
increase tax payment
• Users can donate tax credits to charities
– Charities must type the receipt parameters (40-60 characters
per receipt)
• DoeNota is an App (iOS, Android and Microsoft)
– Scans the receipt (picture)
– Sends to our server (cloud)
– Apply OCR
– Donate to charity through the government site
– 2683 receipts already processed (total value of R$165.694,64)
– Migrated to cloud for scalability in less than a week
Porto Alegre, July 6th, 2016 8

9. New Memory Technologies
• Non-volatile memory as main memory
• New opportunities for changing programming
models
– Data Persistence
– Databases
• New security threats
– Passwords will not be erased if left in memory
• New computer organization
– Universal memory
Porto Alegre, July 6th, 2016 9
ISCA 2013

10. Thank you!
Rodolfo Azevedo
rodolfo@ic.unicamp.br