Workshop Program (1st October 2019)

  • 08:45 - 09:00 Welcome and Opening

  • 09:00 - 10:00 Keynote by Peter Van Roy (UC Louvain)
    Title: Reflections on scalability and consistency

    Abstract:
    Building scalable systems is a neverending quest, because at each higher scale new problems appear that require new ideas to solve. For the Internet we may eventually reach a final solution, when the Internet’s size approaches its final limit. However, the Internet has been growing exponentially since its inception in 1970, and today that growth continues as strong as ever, manifest in the exponential growth of Internet of Things. So understanding scalability is just as important today as it was in the early days. In this talk, I will give a personal overview of some important principles that govern scalable systems. Some will be well-known to you, and (I hope) some will be novel. I will first present some unusual phenomena that appear at large scales, such as burstiness (Heisenberg applications), Buridan’s problem, black swans, and fragility due to network effects (giant components, cascades, etc.). Then I will give some ideas on scalable management principles to solve these and other problems, starting with autonomic computing, leading to feedback structures and convergent data management. During the talk I will make connections with well-known tools such as the throughput-latency diagram and the CAP theorem.

  • 10:00 - 10:30 Coffee Break

  • 10:30 - 11:00 Keynote by Carlos Baquero (Universidade do Minho)
    Title: Efficient Synchronization of State-based CRDTs

    Abstract:
    To ensure high availability in large scale distributed systems, Conflict-free Replicated Data Types (CRDTs) relax consistency by allowing immediate query and update operations at the local replica, with no need for remote synchronization. State-based CRDTs synchronize replicas by periodically sending their full state to other replicas, which can become extremely costly as the CRDT state grows. Delta-based CRDTs address this problem by producing small incremental states (deltas) to be used in synchronization instead of the full state. However, current synchronisation algorithms for Delta-based CRDTs induce redundant wasteful delta propagation, performing worse than expected, and surprisingly, no better than State-based. In this talk we: 1) identify two sources of inefficiency in current synchronization algorithms for delta-based CRDTs; 2) bring the concept of join decomposition to state-based CRDTs; 3) exploit join decompositions to obtain optimal deltas and 4) improve the efficiency of synchronization algorithms; and finally, 5) evaluate the improved algorithms.

  • 11:00 - 11:30 Evaluation and Ranking of Replica Deployments in Geographic State Machine Replication. Shota Numakura, Junya Nakamura and Ren Ohmura.

  • 11:30 - 12:00 Keynote by Brad King (Scality)
    Title: Do you really need a distributed system?

    Abstract:
    Distributed systems approaches have gotten a lot of press as a scaling method for platforms servicing multi-million-user customer-bases. Systems of a planetary scale, such as those deployed by global Internet companies, seem to have demonstrated the effectiveness of horizontal distributed scaling. However, these advantages come with costs and this presentation will discuss some real-world situations to determine to what extent a distributed approach is really worth the effort. Centralized storage of state is invariably simpler than distributed storage of state, but centralized approaches may meet limits that cannot be overcome. For example, flash-based storage-systems as well as modern key-value databases with reduced transactional guarantees scale to higher transactional rates, but limits can still be exceeded. In the end, are distributed systems approaches to scaling large platforms just a complex solution looking for a problem or are they the best way forward? Scality has deployed a number of large-scale mission-critical object-based storage systems around the world, and this presentation will respond, at least in part, to the question of the need for distributed storage platforms at large scale.

  • 12:00 Closing Remarks and Lunch