Conversation

Notices

1. Embed this notice
   therealdatabass (ifesdjeen@hachyderm.io)'s status on Friday, 09-Dec-2022 10:06:42 JST therealdatabass

   Dec 4: This paper dives into the notion of time in the distributed system and explores what advantages we can get from relaxing commonly demanded consistency levels, allowing to achieve PRAM, and even causal consistency without sacrifice to availability (i.e. unidirectional streaming if just one node is reachable, making local reads consistent). This can be useful for federating services and stores: "Continuous Integration of Data Histories into Consistent Namespaces"

   https://arxiv.org/pdf/2204.00470.pdf

   • Embed this notice
     therealdatabass (ifesdjeen@hachyderm.io)'s status on Friday, 09-Dec-2022 10:06:43 JST therealdatabass

     in reply to

     Dec 8: Building on top of Flexible Paxos (which Accord also references in quorum reconfig section), Relaxed Paxos explores ways to optimise further than FP (which states that only intersections between the leader election & replication phases are required, not between all quorums), and skip a round when deciding on the already-proposed values. These results are very useful for ones implementing Paxos in the industry: "Relaxed Paxos: Quorum Intersection Revisited (Again)"

     https://arxiv.org/pdf/2203.03058.pdf

   • Embed this notice
     therealdatabass (ifesdjeen@hachyderm.io)'s status on Friday, 09-Dec-2022 10:06:45 JST therealdatabass

     in reply to

     Dec 7: In addition to modelling database behaviour, it is important to produce relevant and realistic testing scenarios. System upgrades are generally something overlooked during testing, which means that upgrades are going to be causing most operational pain. This paper introduces Distributed System Upgrade Tester, and explores upgrade failures in a number of distributed systems: "Understanding and Detecting Software Upgrade Failures in Distributed Systems"

     https://dl.acm.org/doi/10.1145/3477132.3483577

     Kris Nóva repeated this.
   • Embed this notice
     therealdatabass (ifesdjeen@hachyderm.io)'s status on Friday, 09-Dec-2022 10:06:46 JST therealdatabass

     in reply to

     Dec 6: A good property-based test requires a good model, and modelling complex properties, such as linearizability and strict serializability is not always that simple, both in terms of time complexity and ease of understanding. Elle shows how to reframe the problem and make the operation results help with dependency tracking: "Elle: Inferring Isolation Anomalies from Experimental
     Observations"

     https://people.ucsc.edu/~palvaro/elle_vldb21.pdf

   • Embed this notice
     therealdatabass (ifesdjeen@hachyderm.io)'s status on Friday, 09-Dec-2022 10:06:47 JST therealdatabass

     in reply to

     Dec 5: Most of the time you hear about general purpose indexes that require natural comparison, and tries are most often mentioned and used in conjunction with strings. This paper discusses byte-ordered types and general purpose trie-based indexes that you can use with any data type (of course, preserving the order): “True Memtables in Cassandra”

     https://www.vldb.org/pvldb/vol15/p3359-lambov.pdf