- Surprising Scalability of Multitenancy: Work load packing, non-correlated workloads, and peak vs average utilization. Some important insights from working around s3, ec2 and other large systems about the benefits of many users and different workloads to system utilization. There is a risk of correlated peaks that has to be managed.
- Building and Operating a Pretty Big Storage System (My Adventures in Amazon S3): This is the video about s3 Marc speaks to above. With multi-tenancy, peaks are a bit more peaky, but means that start out as periods of idle time fill in!
- Heat management across a service (hot spots, areas of large activity)
- Looks like queuing, stalling, tail latency
- Objects are spread across many disks so that IO can come from many more drive sources
- There exists a customer with data in s3 that is stored across a million physical drives
