C6

Question 1

geographical load balancing

Answer 1

• we have multiple data centers available at different locations
• QoS (Quality of Service) parameters given: deadlines and latency requirements
• variable: electricity prices for different regions
  => balance the load for the best price, without violating QoS parameters

Question 2

QoS metrics for IaaS and PaaS/SaaS

Answer 2

IaaS: infrastructure availability and performance
PaaS/SaaS: availability, latency, throughput

agreed service formalized in Service-Level Agreement (SLA)

Question 3

interest of the cloud provider

Answer 3

adhering to agreed QoS, but only by a small margin and optimize for cost (hardware and energy consumption)
=> Given a list of allocated VMs, how to map these on a given set of physical hardware in the most efficient manner?

• too much physical hardware in use: high energy consumption
• too few physical hardware in use: risk of overloading and consequently SLA violations
  QoS vs. energy trade-off

Question 4

which VM to run on which physical host?

Answer 4

need for a scheduler running at data-center level

objectives:
– minimize operational cost, maximize income, avoid SLA penalties
– minimize energy consumption
– minimize migrations, minimize network traffic
– avoid overloaded nodes and perform thermal management by spreading the load

Question 5

when to run the scheduler

Answer 5

• when new resources are requested or old ones are released
• at node failure
• periodically, to see whether we can optimize for the current situation

Question 6

pricing models in use today

Answer 6

• on-demand: you need a VM right now and pay the current price
• spot instances: when there’s a lot of available capacity, this can be rented at a reduced price, or bidding (user specifies until what price to keep the instance and pay current price)
• leasing for a set duration: you enter an agreement to pay for a certain period. In return, you get a discounted price.