Software architectures and tools for highly distributed applications Voldemaras Žitkus
Direction Shared nothing architecture. MapReduce-like frameworks. Dynamic reconfiguration of applications. Cloud computing. 2
Shared nothing architecture Used in NoSQL, Data warehouses, Big Data. Solutions:Teradata, Oracle Database Machine, Netezza, DATAllegro. 3
Scalability and elasticity Scalability – capability of an application to increase its total output under an increased load when resources are added. Elasticity – capability of an application to adapt to workload in real time. 4
Dynamic reconfiguration 5
MapReduce framework 6
Heterogeneous cloud computing Multiple providers. Higher computational capabilities. Lower cost. Easier to switch providers. 7
Challenges of heterogeneous clouds Data location. Data throughput. Latency. “Slowest node” problem. 8
Data location and audit (1) R1: Completeness, all infrastructures level VM operations must be logged. R2: Integrity of all monitored events and the log must be guaranteed. R3: Federation, all application events must be recorded even in the federation. R4: Loosely coupled federation, monitoring must be performed in a loosely coupled federation 9
Data location and audit (2) R5: Monitored, all VM operations must be monitorable by the SP. R6: Flexible required log format, LOG per client: (VM, Operation, Time, VEEM certificate) R7: Isolation, a given SP only has access to monitoring information on his own VM, i.e. the ones he has deployed on the cloud. R8: data location, the physical location of data must be transparent, i.e. the physical location of a customer’s data can easily be known. 10
Data throughput Different approaches for different problems. Capability to operate on encrypted data. 11
Latency Dedicated connectivity to cloud providers: AWS Direct Connect, Windows Azure. Cloud Hubs: Interxion. Not all applications require low latency. 12
“Slowest node” problem Redundant execution. Monitoring and migrating resource- constrained virtual machines. 13
Hardware Shift from commodity hardware to specialized. New hyperscale servers: IBM‘s NeXtScale. Extreme low-energy servers: HP‘s Moonshot. 14
Questions 1.Benefits of shared-nothing architecture? 2.Difference between scalability and elasticity? 3.What is heterogeneous cloud? 4.Main challenges of the heterogeneous clouds? 5.Why data location in the cloud is important? 15