Bacalhau project report 20220211

Another great week of progress.

• We deployed Bacalhau "in production" on a 3 node cluster with public IP addresses and alongside IPFS daemons to test IPFS routing with public IPs - this works great
• A new bacalhau results list <job_id> subcommand to show links to the results and locally cached results and a calculated difference score
• A new bacalhau results fetch <job_id> subcommand to fetch the results to the local node
• The calculated difference score - it compares the traces to identify outliers (likely lying or errant compute nodes). The traces operate on the CPU and memory data and calculate a series of waypoints and just look at the error between each result and the mean value for CPU and memory usage at each waypoint. A low difference score means that a result is similar to the majority. A high difference score means that the result is likely an outliar. This is a simple initial approach but can be extended with lots of signal processing cleverness we can add later. We also plan to compare syscall fingerprints and output data entropy.

Demo! https://filecoinproject.slack.com/archives/C02RLM3JHUY/p1644600223304689