2.5 Resource management, provisioning and scheduling

Work through these

• Resources in the cloud: compute, memory, storage, network, and the quota that limits each

  Compute, memory, storage and network are the four things being allocated, and each carries a quota you can hit without warning. Knowing your quotas is part of knowing your capacity.

• Provisioning methods: static, dynamic, and user self-provisioning — with the over-provisioning trap

  Capacity can be assigned up front, adjusted automatically, or requested by users themselves, and the last of those is where the over-provisioning habit comes from. Everyone asks for more than they need when asking is free.

• Resource scheduling and placement; bin packing, affinity and anti-affinity rules

  Scheduling decides which of your workloads lands on which machine, and the rules let you pack things tightly or deliberately keep them apart. Anti-affinity is what stops two copies of the same service dying together.

• Inter-cloud resource management, and the HaaS / LaaS / NaaS vocabulary you will meet in textbooks

  Managing resources across more than one cloud brings its own vocabulary, and textbooks name hardware, licensing and network as services in their own right. Recognising those terms keeps older material readable.

• Elasticity measured: how fast can you actually get a hundred more machines?

  Elasticity is worth measuring rather than assuming: how long a hundred extra machines actually take to arrive determines whether autoscaling can absorb your spike. The number is different for every service.