"Enso: A Streaming Interface for NIC-Application Communication" https://www.microsoft.com/en-us/research/uploads/prod/2023/0...

We've been using the same API to communicate with our NICs since 1994. That API severely limits network throughput and latency. By simply changing the API (no new NIC) you can get 6x higher throughput in some apps and 43% lower latency.

Code runs on FPGA NIC only for now: https://github.com/crossroadsfpga/enso

Won USENIX OSDI best paper award and best artifact award.

This is the best systems paper of the year so far imo

This is also a good (applied, with simple code) example of the use of probabilistic programming. I can't get myself to read full books, but somehow this simple example gave me some intuition and additional pointers to follow.

For everyone else who may be similarly confused as I, due to associating this with Golang or the direction of a startup: take this title literally.

Subtitle is "Neuroscientists discover a mechanism for brain-wide communication when selecting a route toward a destination."

I really support this and am an avid cyclist myself. But this really only works in temperate (and dry) climates. If it rains, we'll have a hard time to convince people to use something without a build-in roof.

It turns out that most caching algorithms we use in practice (CDNs, memcached, storage, etc.) are still pretty far from the optimal hit ratio/ miss ratio.

See, e.g., recent work that exposes that gap (I'm an author)

https://www.cs.cmu.edu/~dberger1/pdf/2018PracticalBound_SIGM...

https://github.com/dasebe/optimalwebcaching

Yes, that paper LHD (from a colleague of mine) does use online stochastic modeling of the workload to make better decisions. So this definitely fits the bill

Author here: thank you pointing out the connection to parallel DBMs - I was not aware of the problem similarity but difference in approach.

Can you point me to a DBMs that does that, both an OSS or in production?

Author here: this is very much a research project with the purpose of prototyping the idea that caching can help tail latency. Previously, people didn't believe that caches can be used that way and it turns out you really have to rethink how you use the caches.

We build a testbed and replayed production workloads. This is not running in production yet. We're actively looking for new workloads and scenarios to test this at scale!

Direct link to paper: https://www.usenix.org/system/files/osdi18-berger.pdf

Repo: https://github.com/dasebe/robinhoodcache

What's the license of RobinHood's core algorithm? This seems useful.

But this being sponsored research (Microsoft) brings me to thinking of the old ARC cache licensing problem.

http://www.varlena.com/GeneralBits/96.php

No, we do not expect any licensing problems. There's an open issue on github: https://github.com/dasebe/robinhoodcache/issues/1

The issue mentions 3 Clause BSD, which will probably be implemented today.

Haven't tested this, and it's a true concern. But I would hope there's some truth to emission reductions as having a comprehensive pass (includes bikes, buses, and Helsinkis good subway system) should entice people to use these options as well. One incentive to use these other options is not needing to search for a parking spot. To make a real difference, people need to be willing to not rely on car sharing 100%, though. So maybe more incentives would be needed for this to reduce emissions in other cities.