High-level benchmarking suite, as part of attempt to speed up Tahoe

[itamarst]

We should have a broader benchmarking attempt, involving:

• Defining metrics
• Defining goals

Two aspects:

• User experience
• Server-side load

UX is probably more important at this point, but eventually if that's finished server load will become an issue.

Metrics:

• Latency (single upload/download)
• Throughput (batch uploads/downloads)

Bottlenecks will be network latency, CPU, network bandwidth. If we've hit point where network bandwidth is the bottleneck, we're done optimizing.

Throughput depends on usage patterns of:

• tahoe cp
• tahoe backup
• sftp
• Magic Folders
• Third party backup tools that integrate with Tahoe (duplicaty? duplicati? borg?)

Knobs: number of backend storage servers, network latency.

[itamarst]

Might want to get ​https://github.com/tahoe-lafs/tahoe-lafs/pull/1208 merged and then rely on the code that was refactored out of the tests.