Ah, thanks. How much of it is attributable to BEAM itself, I wonder. Using AOT instead of the JIT cuts memory usage in .NET by about half, for ex. IIRC, the numbers below are steady state under load, not startup, but either way

Pat Helland delivered a presentation on this paper: hpts.ws/papers/2024/...

all threads are cranking away to test the impl

Description: github.com/dotnet/orlea...
Proposal: github.com/dotnet/orlea...
Implementation: github.com/dotnet/orlea...
I'm not sure that it's necessarily "Virtual Synchrony" rather than merely inspired by Virtual Synchrony - if you read it, let me know what you think

Consistent membership greatly simplifies the design of distributed algorithms. Eg, Orleans's grain directory employs Virtual Synchrony, which requires consistent membership views. RAPID has a neat application of Fast Paxos for agreeing on membership views. Anyway, read the paper & welcome Lalith!

Distributed Systems & #DataBS folks, please welcome @lalithsuresh.bsky.social! Lalith authored my favorite paper on group membership, RAPID. Instead of building strongly consistent membership atop weak (eg, Raft + SWIM), RAPID delivers consistency from the get-go www.usenix.org/system/files...

SQL Server Hyperscale also uses the disaggregated model. For details, see the paper: Socrates: The New SQL Server in the Cloud www.microsoft.com/en-us/resear...

Maybe it's too fiddly to use in practice. The person in that SO thread seemed to get good results with it in the end

Scan RAM without trashing your CPU caches by leveraging non-temporal instructions, Memory Type Range Registers, or cache line flushing. Obscure, but perhaps useful for minimizing the impact of garbage collectors / LSM compaction. dl.acm.org/doi/pdf/10.1...

Here's an example of durable futures. In this case, intermixed with stateful objects (grains). This PoC is backed by object storage: each grain has a reliable state machine under the hood, but the interface is more like a typical object.

Disaggregated storage pushes us to reexamine past ideas under a new light. Phil Bernstein gives middle-tier distributed transactions another look in this paper sites.computer.org/debull/A19ju.... What other examples come to mind?

The Vertical Paxos paper describes some of the challenges of using consensus for data replication and concedes that consensus is often used only for configuration, with other mechanisms for replicating the data itself. The introduction, at least, is worth reading: www.microsoft.com/en-us/resear...

Orleans can automatically migrate frequently communicating grains to the same hosts, reducing network calls. Here's an app running on a real cluster showing grains (small circles) being migrated between hosts to eliminate network calls (red lines).

A new feature in Orleans enables it to seamlessly balance already-active grains across your cluster as it scales or rolls. In this demo, 100K grains are activated on a single node. The cluster is scaled to 10 nodes and the grains are gradually spread across them until everything is balanced.