Isolation vs. utilization
Where you meet it: multi-tenancy, resource pools, spot capacity.
What it is
Three tenants have peak loads of 100, 100 and 100 units, but never at the same time; the combined peak is 150. Hard partitions need 300 units. Sharing needs 150. Isolation costs twice as much.
From Emergent Failure Dynamics › Resource exhaustion
Between reserving a resource up front and taking it back when it is needed, there is no third option. Where you stand on this line is a statement about how strongly you believe your workload's peaks are correlated (happen at the same time). The recurring problem here is that independence is assumed but false: components are assumed to peak at different times, and they usually peak together. Monday at 9 am is the example.
From Memory & Resource Management › Taking a resource back
The isolation ladder is familiar: shared process, shared cluster with separate processes, separate cluster, separate account. The cost is the isolation-versus-utilization axis, paid in money (the more you separate, the more idle capacity you pay for). The rule: anything shared between tenants is a channel between tenants, including things that carry only timing.
From Security & Privacy › Egress and tenancy
Where it appears
In 7 mechanisms. Each says what the mechanism is about, then how this trade-off shows up there, in its own words.
- Memory & Resource Management
Who is allowed to hold how much memory, for how long, and what happens to the work when the answer is less than it asked for?
The structural lesson reframes this whole article: memory is two resources, not one.
Read it inCapacity isolation is not bandwidth isolationWhat it costs to take a resource back
- Caching
What happens when we deliberately keep a second copy of data that no protocol forces to be correct?
The last row shows that the guarantee was bought with the elasticity that was the point of a cache.
Read it inPartitioned capacity: four arguments, one structureKnees and plateaus in the miss-ratio curve
- Emergent Failure Dynamics
What happens when nothing is broken, every component is healthy, and the system's own response to stress makes the stress worse?
One line: a shared pool converts one dependency's latency into every dependency's unavailability, on a clock of P/λ seconds.
Read it inBlast radius, failure domains and cellsResource-exhaustion cascade
- Scheduling & Workload Management
Which work runs now, which work waits, and who pays for the waiting?
Fairness leaves resources stranded that a bin-packer would use (the 4 GB above).
Read it inFair and capacity schedulingAnti-affinityBin packing versus spread placementThree questions, one tension, and conserved workPrioritization and short-query lanesWorkload isolation: pools, warehouses, clusters
- Cost Mechanics
What is metered, and what is the smallest increment you are charged for?
Attribution granularity cannot exceed isolation granularity.
Read it inTagging and attributionServerless pricing and warm poolsCompute: capacity-time, not work
- Streaming
What does it cost to remember, when the input never ends?
Queryable state trades every isolation property for one network hop.
Read it inQueryable state
- Measurement & Experimental Method
What would have to be true for this number to mean what I think it means?
Realism is bought with exposure, and shared infrastructure makes the two arms contaminate each other.
Related
Made on the same pages most often.