Presentation is loading. Please wait.

Presentation is loading. Please wait.

Towards a dynamic multi-cloud computing universe Divy Agrawal & Amr El Abbadi UC Santa Barbara

Similar presentations


Presentation on theme: "Towards a dynamic multi-cloud computing universe Divy Agrawal & Amr El Abbadi UC Santa Barbara"— Presentation transcript:

1 Towards a dynamic multi-cloud computing universe Divy Agrawal & Amr El Abbadi UC Santa Barbara amr@cs.ucsb.edu

2 Origins of Cloud Single data center Focus: – Virtualization technologies – Reliable and robust infrastructure What if the data center fails? Source: http://www.logichp.com/2010/04/21/hps-houston-data- center-cisco-free-twice-as-efficient/

3 Cloud Evolution Multiple static data centers Geo-replication – Fault-tolerance – Lower latency across regions What are we missing? Source: http://www.clusterdb.com/wp- content/uploads/2009/08/multi_master_replication2.jpg

4 Cloud Evolution Too many providers – Potential for vendor lock-in – Lack of cloud standards Focus only on large “cores” – Missing out on resources at the edge Source: http://www.valuecdn.com/images/cdn-globe.jpg Source: http://fennelway.com/img/cloud-lock.jpg

5 Cloud Revolution Build a cloud ecosystem: Lots of resources at the edge – Content Delivery Networks Huge potential with transient resources – Wall street companies – Large infrastructures with diurnal usage Inter-operability between cloud providers

6 Dynamic Clouds (diurnal behavior) 1:00 am 9:00 am Transition through time epochs Nucleus

7 Research Vision Design Principles: – Separate system and application data – Decouple control from storage of date A collection of static cloud “nucleus” – Heart and mind: coordinate and synchronize. A collection of dynamic cloud “cores” – Augment system capacity.

8 Technical Challenges A uniform namespace for multi cloud cores. Efficient integration of surplus capacity. Effective load and data migration. Scalable monitoring and system modeling.

9 Agility and Elasticity Managing Churn – Predictable; checkpoints. Load balancing and elasticity – Migration; replication Data Placement – Need models to account for churn and load changes.

10 Building a Dynamic Cloud System Monitoring and Control – System stats, failures, data placement, etc Metadata Management (system state) – Strong consistency within a single core – Casual consistency across cores (dist dictionary)? Data Access Control – Inexpensive tenant migration Data Storage (fault tolerance) – Cross core replication—timeline consistency?

11 More Issues Standards Differential pricing: varying demand and availability SLAs versus Operating cost Privacy and security Supporting diverse applications and workloads


Download ppt "Towards a dynamic multi-cloud computing universe Divy Agrawal & Amr El Abbadi UC Santa Barbara"

Similar presentations


Ads by Google