Current serverless offerings give users limited flexibility for configuring the resources allocated to their function invocations. This simplifies the interface for users to deploy server-less computations but creates deployments that are resource inefficient. In this paper, we take a principled approach to the problem of resource allocation for serverless functions, analyzing the effects of automating this choice in a way that leads to the best combination of performance and cost. In particular, we systematically explore the opportunities that come with decoupling memory and CPU resource allocations and also enabling the use of different VM types, and we find a rich trade-off space between performance and cost. The provider can use this in a number of ways, e.g., exposing all these parameters to the user; eliding preferences for performance and cost from users and simply offer the same performance with lower cost; or exposing a small number of choices for users to trade performance for cost. Our results show that, by decoupling memory and CPU allocation, there is the potential to have up to 40% lower execution cost than the preset coupled configurations that are the norm in current serverless offerings. Similarly, making the correct choice of VM instance type can provide up to 50% better execution time. Furthermore, we demonstrate that providers have the flexibility to choose different instance types for the same functions to maximize resource utilization while providing performance within 10--20% of the best resource configuration for each respective function.
1 December 2018 in Proceedings of the VLDB Endowment
It is common to use weakly consistent replication to achieve high availability and low latency at a global scale. In this setting, concurrent updates may lead to states where application invariants do not hold. Some systems coordinate the execution of (conflicting) operations to avoid invariant violations, leading to high latency and reduced availability for those operations. This problem is worsened by the difficulty in identifying precisely which operations conflict. In this paper we propose a novel approach to preserve application invariants without coordinating the execution of operations. The approach consists of modifying operations in a way that application invariants are maintained in the presence of concurrent updates. When no conflicting updates occur, the modified operations present their original semantics. Otherwise, we use sensible and deterministic conflict resolution policies that preserve the invariants of the application. To implement this approach, we developed a static analysis, IPA, that identifies conflicting operations and proposes the necessary modifications to operations. Our analysis shows that IPA can avoid invariant violations in many applications, including typical database applications. Our evaluation reveals that the offline static analysis runs fast enough for being used with large applications. The overhead introduced in the modified operations is low and it leads to lower latency and higher throughput when compared with other approaches that enforce invariants.
One of the most recent members of the Paxos family of protocols is Generalized Paxos. This variant of Paxos has the characteristic that it departs from the original specification of consensus, allowing for a weaker safety condition where different processes can have a different views on a sequence being agreed upon. However, much like the original Paxos counterpart, Generalized Paxos does not have a simple implementation. Furthermore, with the recent practical adoption of Byzantine fault tolerant protocols in the context of blockchain protocols, it is timely and important to understand how Generalized Paxos can be implemented in the Byzantine model. In this paper, we make two main contributions. First, we attempt to provide a simpler description of Generalized Paxos, based on a simpler specification and the pseudocode for a solution that can be readily implemented. Second, we extend the protocol to the Byzantine fault model, and provide the respective correctness proof.
Most geo-replicated storage systems use weak consistency to avoid the performance penalty of coordinating replicas in different data centers. This departure from strong semantics poses problems to application programmers, who need to address the anomalies enabled by weak consistency. In this paper we use a recently proposed isolation level, called Non-Monotonic Snapshot Isolation, to achieve ACID transactions with low latency. To this end, we present Blotter, a geo-replicated system that leverages these semantics in the design of a new concurrency control protocol that leaves a small amount of local state during reads to make commits more efficient, which is combined with a configuration of Paxos that is tailored for good performance in wide area settings. Read operations always run on the local data center, and update transactions complete in a small number of message steps to a subset of the replicas. We implemented Blotter as an extension to Cassandra. Our experimental evaluation shows that Blotter has a small overhead at the data center scale, and performs better across data centers when compared with our implementations of the core Spanner protocol and of Snapshot Isolation on the same codebase.