some scientific challenges in the cloud - roberto di cosmo
DESCRIPTION
TRANSCRIPT
Some Scientific Challenges in the Cloud
Roberto Di CosmoUniversity Paris Diderot, and INRIA
May 20, 2010GTLLParis
Roberto Di Cosmo () Scientific challenges in the Cloud May 20 2010 / Paris 1 / 5
Cloud Computing: basic definitionsThree levels:
Layer Main target Abstraction
Software as a Service (SaaS) Final users Application
Platform as a Service (PaaS) App Developers System
Infrastructure as a Service (IaaS) Platform Developers HW
Key concepts: virtualization + flexibility + automation
Erich Clementi, IBM’s Cloud Initiative head says it well:
Many people equate cloud computing to virtualization.It is not virtualization.To get the value you need standardization and automation ontop of that virtualization.
Let’s sketch some scientific challenges at each level.
Roberto Di Cosmo () Scientific challenges in the Cloud May 20 2010 / Paris 2 / 5
Cloud Computing: basic definitionsThree levels:
Layer Main target Abstraction
Software as a Service (SaaS) Final users Application
Platform as a Service (PaaS) App Developers System
Infrastructure as a Service (IaaS) Platform Developers HW
Key concepts: virtualization
+ flexibility + automation
Erich Clementi, IBM’s Cloud Initiative head says it well:
Many people equate cloud computing to virtualization.It is not virtualization.To get the value you need standardization and automation ontop of that virtualization.
Let’s sketch some scientific challenges at each level.
Roberto Di Cosmo () Scientific challenges in the Cloud May 20 2010 / Paris 2 / 5
Cloud Computing: basic definitionsThree levels:
Layer Main target Abstraction
Software as a Service (SaaS) Final users Application
Platform as a Service (PaaS) App Developers System
Infrastructure as a Service (IaaS) Platform Developers HW
Key concepts: virtualization + flexibility
+ automation
Erich Clementi, IBM’s Cloud Initiative head says it well:
Many people equate cloud computing to virtualization.It is not virtualization.To get the value you need standardization and automation ontop of that virtualization.
Let’s sketch some scientific challenges at each level.
Roberto Di Cosmo () Scientific challenges in the Cloud May 20 2010 / Paris 2 / 5
Cloud Computing: basic definitionsThree levels:
Layer Main target Abstraction
Software as a Service (SaaS) Final users Application
Platform as a Service (PaaS) App Developers System
Infrastructure as a Service (IaaS) Platform Developers HW
Key concepts: virtualization + flexibility + automation
Erich Clementi, IBM’s Cloud Initiative head says it well:
Many people equate cloud computing to virtualization.It is not virtualization.To get the value you need standardization and automation ontop of that virtualization.
Let’s sketch some scientific challenges at each level.
Roberto Di Cosmo () Scientific challenges in the Cloud May 20 2010 / Paris 2 / 5
Cloud Computing: basic definitionsThree levels:
Layer Main target Abstraction
Software as a Service (SaaS) Final users Application
Platform as a Service (PaaS) App Developers System
Infrastructure as a Service (IaaS) Platform Developers HW
Key concepts: virtualization + flexibility + automation
Erich Clementi, IBM’s Cloud Initiative head says it well:
Many people equate cloud computing to virtualization.It is not virtualization.To get the value you need standardization and automation ontop of that virtualization.
Let’s sketch some scientific challenges at each level.
Roberto Di Cosmo () Scientific challenges in the Cloud May 20 2010 / Paris 2 / 5
Cloud Computing: basic definitionsThree levels:
Layer Main target Abstraction
Software as a Service (SaaS) Final users Application
Platform as a Service (PaaS) App Developers System
Infrastructure as a Service (IaaS) Platform Developers HW
Key concepts: virtualization + flexibility + automation
Erich Clementi, IBM’s Cloud Initiative head says it well:
Many people equate cloud computing to virtualization.It is not virtualization.To get the value you need standardization and automation ontop of that virtualization.
Let’s sketch some scientific challenges at each level.Roberto Di Cosmo () Scientific challenges in the Cloud May 20 2010 / Paris 2 / 5
Challenges in IaaS
Optimization of resources:
share common data in disk and memory (see DecentralizedDeduplication in SAN Cluster File Systems, Clements et al., andSatori: Enlightened page sharing, Milos et al. USENIX ’09)
optimal VM placement
Security:
avoiding escapes
avoiding mapping of the physical infrastructure, and its exploitationhttp://cseweb.ucsd.edu/~hovav/dist/cloudsec.pdf
Roberto Di Cosmo () Scientific challenges in the Cloud May 20 2010 / Paris 3 / 5
Challenges in IaaS
Optimization of resources:
share common data in disk and memory (see DecentralizedDeduplication in SAN Cluster File Systems, Clements et al., andSatori: Enlightened page sharing, Milos et al. USENIX ’09)
optimal VM placement
Security:
avoiding escapes
avoiding mapping of the physical infrastructure, and its exploitationhttp://cseweb.ucsd.edu/~hovav/dist/cloudsec.pdf
Roberto Di Cosmo () Scientific challenges in the Cloud May 20 2010 / Paris 3 / 5
Challenges in IaaS
Optimization of resources:
share common data in disk and memory (see DecentralizedDeduplication in SAN Cluster File Systems, Clements et al., andSatori: Enlightened page sharing, Milos et al. USENIX ’09)
optimal VM placement
Security:
avoiding escapes
avoiding mapping of the physical infrastructure, and its exploitationhttp://cseweb.ucsd.edu/~hovav/dist/cloudsec.pdf
Roberto Di Cosmo () Scientific challenges in the Cloud May 20 2010 / Paris 3 / 5
Challenges in IaaS
Optimization of resources:
share common data in disk and memory (see DecentralizedDeduplication in SAN Cluster File Systems, Clements et al., andSatori: Enlightened page sharing, Milos et al. USENIX ’09)
optimal VM placement
Security:
avoiding escapes
avoiding mapping of the physical infrastructure, and its exploitationhttp://cseweb.ucsd.edu/~hovav/dist/cloudsec.pdf
Roberto Di Cosmo () Scientific challenges in the Cloud May 20 2010 / Paris 3 / 5
Challenges in IaaS
Optimization of resources:
share common data in disk and memory (see DecentralizedDeduplication in SAN Cluster File Systems, Clements et al., andSatori: Enlightened page sharing, Milos et al. USENIX ’09)
optimal VM placement
Security:
avoiding escapes
avoiding mapping of the physical infrastructure, and its exploitationhttp://cseweb.ucsd.edu/~hovav/dist/cloudsec.pdf
Roberto Di Cosmo () Scientific challenges in the Cloud May 20 2010 / Paris 3 / 5
Challenges in PaaS
Ensuring flexibility, avoiding vendor lock-in, is not easy.
N.B.: (re)writing applications for a platform/middleware is not neutral.
Roberto Di Cosmo () Scientific challenges in the Cloud May 20 2010 / Paris 4 / 5
Challenges in PaaS
Ensuring flexibility, avoiding vendor lock-in, is not easy.
N.B.: (re)writing applications for a platform/middleware is not neutral.
Roberto Di Cosmo () Scientific challenges in the Cloud May 20 2010 / Paris 4 / 5
Challenges in PaaS
Ensuring flexibility, avoiding vendor lock-in, is not easy.
N.B.: (re)writing applications for a platform/middleware is not neutral.
Roberto Di Cosmo () Scientific challenges in the Cloud May 20 2010 / Paris 4 / 5
Transversal Challenges
Several challenges are transversal:
security:who can break (into) the system?
data ownership:who can access my data? please do not simply rely on contracts!
efficient implementation:abstraction is nice, but can it accomodate optimal implementation?
interoperability:SaaS and PaaS try to hide the complexity of the lower layer, but atthe price of vendor or technology lock-in
Bottomline: there is a lot of work ahead.
Roberto Di Cosmo () Scientific challenges in the Cloud May 20 2010 / Paris 5 / 5
Transversal Challenges
Several challenges are transversal:
security:who can break (into) the system?
data ownership:who can access my data?
please do not simply rely on contracts!
efficient implementation:abstraction is nice, but can it accomodate optimal implementation?
interoperability:SaaS and PaaS try to hide the complexity of the lower layer, but atthe price of vendor or technology lock-in
Bottomline: there is a lot of work ahead.
Roberto Di Cosmo () Scientific challenges in the Cloud May 20 2010 / Paris 5 / 5
Transversal Challenges
Several challenges are transversal:
security:who can break (into) the system?
data ownership:who can access my data? please do not simply rely on contracts!
efficient implementation:abstraction is nice, but can it accomodate optimal implementation?
interoperability:SaaS and PaaS try to hide the complexity of the lower layer, but atthe price of vendor or technology lock-in
Bottomline: there is a lot of work ahead.
Roberto Di Cosmo () Scientific challenges in the Cloud May 20 2010 / Paris 5 / 5
Transversal Challenges
Several challenges are transversal:
security:who can break (into) the system?
data ownership:who can access my data? please do not simply rely on contracts!
efficient implementation:abstraction is nice, but can it accomodate optimal implementation?
interoperability:SaaS and PaaS try to hide the complexity of the lower layer, but atthe price of vendor or technology lock-in
Bottomline: there is a lot of work ahead.
Roberto Di Cosmo () Scientific challenges in the Cloud May 20 2010 / Paris 5 / 5
Transversal Challenges
Several challenges are transversal:
security:who can break (into) the system?
data ownership:who can access my data? please do not simply rely on contracts!
efficient implementation:abstraction is nice, but can it accomodate optimal implementation?
interoperability:SaaS and PaaS try to hide the complexity of the lower layer, but atthe price of vendor or technology lock-in
Bottomline: there is a lot of work ahead.
Roberto Di Cosmo () Scientific challenges in the Cloud May 20 2010 / Paris 5 / 5
Transversal Challenges
Several challenges are transversal:
security:who can break (into) the system?
data ownership:who can access my data? please do not simply rely on contracts!
efficient implementation:abstraction is nice, but can it accomodate optimal implementation?
interoperability:SaaS and PaaS try to hide the complexity of the lower layer, but atthe price of vendor or technology lock-in
Bottomline: there is a lot of work ahead.
Roberto Di Cosmo () Scientific challenges in the Cloud May 20 2010 / Paris 5 / 5