Aleksey Charapko
@AlekseyCharapko
Followers
3K
Following
791
Statuses
542
Assistant professor of computer science @UofNH. ex-Microsoft. Interested in everything distributed/cloud/edge
Joined November 2011
RT @MarcJBrooker: In this month's ACM Queue, @ankushpd and I write about some of the methods and tools we apply to systems correctness at A…
0
46
0
We are resuming the distributed systems reading group after the winter break with a fresh set of papers for the spring semester. We meet almost every Thursday at 1pm ET over zoom. Schedule, papers and more info on joining:
2
15
100
RT @jessejiryudavis: My review of SwiftPaxos, a complex consensus protocol that minimizes network latency in geo-distributed systems: https…
0
4
0
RT @tevfikkosar: A revised and extended version of our work on “How to Evaluate Distributed Coordination Systems? – A Survey and Analysis”…
0
16
0
Thanks, Marc!
Great choice by the @NSF to fund @AlekseyCharapko and team’s work on metastable failures. Understanding the dynamics of distributed systems is super important to industry, and a fascinating research problem.
0
0
14
Just in for the new year! I have received an NSF CAREER award on the metastable failures project! So excited to see where we will take this in the next 5 years!
4
2
58
RT @MarcJBrooker: My third talk at re:Invent this year was about patterns for system resilience. Here, I talk about the trouble with retrie…
0
34
0
RT @jessejiryudavis: We're experimenting with predictive auto-scaling at MongoDB, here's a talk and writeup about forecasting database dema…
0
16
0
@eatonphil For what it is worth, I think MultiPaxos is easier to understand than Raft, especially if you start to consider cornercases with leader elections that require things like prevote.
0
1
5
@yassinm_tw @owenhilyard And I have not mentioned hardware offloads/acceleration yet, which again is HW-dependent and must hide behind powerful abstractions to be portable.
0
0
0
@muratdemirbas @eatonphil @nvartolomei The problem with MP for replicating data is the need to recover the data in case of a failure. A smart way to decouple the consensus would have a quorum-based replication that can recover on its own and use consensus for lighter stuff, like ordering already replicated operations
0
0
2