[tahoe-dev] Performance testing

[Vladimir Arseniev]

Our group has created a small test storage grid. Is there a standard set
of benchmarks for testing performance? I've read about Brian Warner's
download tests, and have looked through some of the Buildbot pages. But
I don't get how developers assess grid performance and reliability.
We're particularly interested in measuring each node's performance, and
impacts of inter-node latency.

Currently there are an introducer, a client, and nine storage nodes. All
of the nodes are on a private VPN (more about that below). We're
experimenting with remote storage nodes. So far, we have one Atlas
DS1-RS, and two VPS. We're running the introducer, the client, and the
other six storage nodes locally, with Internet connectivity through
multi-hop VPN. I'll be adding some AWS storage nodes shortly.

I'm running the OpenVPN access server locally, with remote ssh to a VPS
in Malaysia. While we had initially run the access server on AWS, we
decided that keeping access credentials there was too risky. Given that
all private VPN traffic bounces off the access server, however, that
adds considerable latency (and a throughput bottleneck).

We welcome suggestions.