Search

Recent high-performance storage devices have exposed software inefficiencies in existing storage stacks, leading to a new breed of I/O stacks. The newest storage API of the Linux kernel is io-uring. We perform one of the first in-depth studies of io-uring, and compare its performance and dis-/advantages with the established libaio and SPDK APIs. Our key findings reveal that (i) polling design significantly impacts performance; (ii) with enough CPU cores io-uring can deliver performance close to that of SPDK; and (iii) performance scalability over multiple CPU cores and devices requires careful consideration and necessitates a hybrid approach. Last, we provide design guidelines for developers of storage intensive applications.

Machine learning (ML) methods have recently emerged as an effective way to perform automated parameter tuning of databases. State-of-the-art approaches include Bayesian optimization (BO) and reinforcement learning (RL). In this work, we describe our experience when applying these methods to a database not yet studied in this context: FoundationDB. Firstly, we describe the challenges we faced, such as unknown valid ranges of configuration parameters and combinations of parameter values that result in invalid runs, and how we mitigated them. While these issues are typically overlooked, we argue that they are a crucial barrier to the adoption of ML self-tuning techniques in databases, and thus deserve more attention from the research community. Secondly, we present experimental results obtained when tuning FoundationDB using ML methods. Unlike prior work in this domain, we also compare with the simplest of baselines: random search. Our results show that, while BO and RL methods can improve the throughput of FoundationDB by up to 38%, random search is a highly competitive baseline, finding a configuration that is only 4% worse than the, vastly more complex, ML methods. We conclude that future work in this area may want to focus more on randomized, model-free optimization algorithms.

Log-Structured Merge Key-Value stores (LSM KVs) are designed to offer good write performance, by capturing client writes in memory, and only later flushing them to storage. Writes are later compacted into a tree-like data structure on disk to improve read performance and to reduce storage space use. It has been widely documented that compactions severely hamper throughput. Various optimizations have successfully dealt with this problem. These techniques include, among others, rate-limiting flushes and compactions, selecting among compactions for maximum effect, and limiting compactions to the highest level by so-called fragmented LSMs. In this article, we focus on latencies rather than throughput. We first document the fact that LSM KVs exhibit high tail latencies. The techniques that have been proposed for optimizing throughput do not address this issue, and, in fact, in some cases, exacerbate it. The root cause of these high tail latencies is interference between client writes, flushes, and compactions. Another major cause for tail latency is the heterogeneous nature of the workloads in terms of operation mix and item sizes whereby a few more computationally heavy requests slow down the vast majority of smaller requests. We introduce the notion of an Input/Output (I/O) bandwidth scheduler for an LSM-based KV store to reduce tail latency caused by interference of flushing and compactions and by workload heterogeneity. We explore three techniques as part of this I/O scheduler: (1) opportunistically allocating more bandwidth to internal operations during periods of low load, (2) prioritizing flushes and compactions at the lower levels of the tree, and (3) separating client requests by size and by data access path. SILK+ is a new open-source LSM KV that incorporates this notion of an I/O scheduler.

In this article, we introduce TAS (Transactional Auto Scaler), a system for automating the elastic scaling of replicated in-memory transactional data grids, such as NoSQL data stores or Distributed Transactional Memories. Applications of TAS range from online self-optimization of in-production applications to the automatic generation of QoS/cost-driven elastic scaling policies, as well as to support for what-if analysis on the scalability of transactional applications. In this article, we present the key innovation at the core of TAS, namely, a novel performance forecasting methodology that relies on the joint usage of analytical modeling and machine learning. By exploiting these two classically competing approaches in a synergic fashion, TAS achieves the best of the two worlds, namely, high extrapolation power and good accuracy, even when faced with complex workloads deployed over public cloud infrastructures. We demonstrate the accuracy and feasibility of TAS’s performance forecasting methodology via an extensive experimental study based on a fully fledged prototype implementation integrated with a popular open-source in-memory transactional data grid (Red Hat’s Infinispan) and industry-standard benchmarks generating a breadth of heterogeneous workloads.

In this paper we investigate the issue of automatically identifying the "natural" degree of parallelism of an application using software transactional memory (STM), i.e., the workload-specific multiprogramming level that maximizes application's performance. We discuss the importance of adapting the concurrency level in two different scenarios, a shared-memory and a distributed STM infrastructure. We propose and evaluate two alternative self-tuning methodologies, explicitly tailored for the considered scenarios. In shared-memory STM, we show that lightweight, black-box approaches relying solely on on-line exploration can be extremely effective. For distributed STMs , we introduce a novel hybrid approach that combines model-driven performance forecasting techniques and on-line exploration in order to take the best of the two techniques, namely enhancing robustness despite model's inaccuracies, and maximizing convergence speed towards optimum solutions.

It is well known that Parallel Discrete Event Simulation systems may suffer, in terms of delivered performance, from imbalance of the computational load. In case of conservative synchronization we may experience CPU under-utilization and/or excessive communication overhead. On the other hand, for the optimistic paradigm we may even have rollback thrashing effects, with a consequent reduction of the percentage of productive (ie not rolled back) work carried out. This paper presents the design of a global memory management architecture supporting application-transparent migration of simulation objects whose state is scattered across dynamically allocated memory chunks. Our approach is based on a non-intrusive background protocol that provides each instance of the simulation kernel with information on the current mapping of the virtual address space of all the other instances. Dynamic memory requests by the application layer are then locally mapped onto virtual-address ranges that maximize the likelihood of being portable onto the address space of a remote kernel instance. In this way, independently of the load-balancing trigger (or policy), we maximize the likelihood that a desirable migration across a specific couple of kernels can actually take place due to compliance of the corresponding source/destination address spaces. We have integrated the global memory manager within the ROme OpTimistic Simulator (ROOT-Sim), namely a run-time environment based on the optimistic synchronization paradigm which automatically and transparently parallelizes the execution of event-handler-based simulation programs conforming to ANSI-C. Further, we provide a contribution in the direction of widening load-balancing schemes for optimistic simulation systems by defining migration triggers and selection policies for the objects to be migrated on the basis of memory usage patterns. An experimental assessment of the architecture and of memory-oriented load balancing is also provided.

The Transactional Memory (TM) paradigm promises to greatly simplify the development of concurrent applications. This led, over the years, to the creation of a plethora of TM implementations delivering wide ranges of performance across workloads. Yet, no universal implementation fits each and every workload. In fact, the best TM in a given workload can reveal to be disastrous for another one. This forces developers to face the complex task of tuning TM implementations, which significantly hampers their wide adoption. In this paper, we address the challenge of automatically identifying the best TM implementation for a given workload. Our proposed system, ProteusTM, hides behind the TM interface a large library of implementations. Underneath, it leverages a novel multi-dimensional online optimization scheme, combining two popular learning techniques: Collaborative Filtering and Bayesian Optimization. We integrated ProteusTM in GCC and demonstrate its ability to switch between TMs and adapt several configuration parameters (e.g., number of threads). We extensively evaluated ProteusTM, obtaining average performance

Performance modeling is a crucial technique to enable the vision of elastic computing in cloud environments. Conventional approaches to performance modeling rely on two antithetic methodologies: white box modeling, which exploits knowledge on system's internals and capture its dynamics using analytical approaches, and black box techniques, which infer relations among the input and output variables of a system based on the evidences gathered during an initial training phase. In this paper we investigate a technique, which we name Bootstrapping, which aims at reconciling these two methodologies and at compensating the cons of the one with the pros of the other. We analyze the design space of this gray box modeling technique, and identify a number of algorithmic and parametric trade-offs which we evaluate via two realistic case studies, a Key-Value Store and a Total Order Broadcast service.

In-memory (transactional) data stores are recognized as a first-class data management technology for cloud platforms, thanks to their ability to match the elasticity requirements imposed by the pay-as-you-go cost model. On the other hand, defining the well-suited amount of cache servers to be deployed, and the degree of in-memory replication of slices of data, in order to optimize reliability/availability and performance tradeoffs, is far from being a trivial task. Yet, it is an essential aspect of the provisioning process of cloud platforms, given that it has an impact on how well cloud resources are actually exploited. To cope with the issue of determining optimized configurations of cloud in-memory data stores, in this article we present a flexible simulation framework offering skeleton simulation models that can be easily specialized in order to capture the dynamics of diverse data grid systems, such as those related to the specific protocol used to provide data consistency and/or transactional guarantees. Besides its flexibility, another peculiar aspect of the framework lies in that it integrates simulation and machine-learning (black-box) techniques, the latter being essentially used to capture the dynamics of the data-exchange layer (e.g. the message passing layer) across the cache servers. This is a relevant aspect when considering that the actual data-transport/networking infrastructure on top of which the data grid is deployed might be unknown, hence being not feasible to be modeled via white-box (namely purely simulative) approaches. We also provide an extended experimental study aimed at validating instances of simulation models supported by our framework against execution dynamics of real data grid systems deployed on top of either private or public cloud infrastructures., 34 pages

Cloud computing represents a cost-effective paradigm to deploy a wide class of large-scale distributed applications, for which the pay-per-use model combined with automatic resource provisioning promise to reduce the cost of dependability and scalability. However, a key challenge to be addressed to materialize the advantages promised by Cloud computing is the design of effective auto-scaling and self-tuning mechanisms capable of ensuring pre-determined QoS levels at minimum cost in face of changing workload conditions. This is one of the keys goals that are being pursued by the Cloud-TM project, a recent EU project that is developing a novel, self-optimizing transactional data platform for the cloud. In this paper we present the key design choices underlying the development of Cloud-TM's Workload Analyzer (WA), a crucial component of the Cloud-TM platform that is change of three key functionalities: aggregating, filtering and correlating the streams of statistical data gathered from the various nodes of the Cloud-TM platform, building detailed workload profiles of applications deployed on the Cloud-TM platform, characterizing their present and future demands in terms of both logical (i.e. data) and physical (e.g. hardware-related) resources, triggering alerts in presence of violations (or risks of future violations) of pre-determined SLAs. © 2012 IEEE.

One approach to fully exploit the potential of Cloud technologies consists in leveraging on the Autonomic Computing paradigm. It could be exploited in order to put in place reconfiguration strategies spanning the whole protocol stack, starting from the infrastructure and then going up to platform/application level protocols. On the other hand, the very base for the design and development of Cloud oriented Autonomic Managers is represented by monitoring sub-systems, able to provide audit data related to any layer within the stack. In this article we present the approach that has been taken while designing and implementing the monitoring sub-system for the Cloud-TM FP7 project, which is aimed at realizing a self-adapting, Cloud based middleware platform providing transactional data access to generic customer applications. © 2012 Springer-Verlag Berlin Heidelberg.

Classical approaches to performance prediction of computer systems rely on two, typically antithetic, techniques: Machine Learning (ML) and Analytical Modeling (AM).ML undertakes a black-box approach, which typically achieves very good accuracy in regions of the features' space that have been sufficiently explored during the training process, but that has very weak extrapolation power (i.e., poor accuracy in regions for which none, or too few samples are known).Conversely, AM relies on a white-box approach, whose key advantage is that it requires no or minimal training, hence supporting prompt instantiation of the target system's performance model. However, to ensure their tractability, AM-based performance predictors typically rely on simplifying assumptions. Consequently, AM's accuracy is challenged in scenarios not matching such assumptions.This tutorial describes techniques that exploit AM and ML in synergy in order to get the best of the two worlds. It surveys several such hybrid techniques and presents use cases spanning a wide range of application domains.

Classical approaches to performance prediction rely on two, typically antithetic, techniques: Machine Learning (ML) and Analytical Modeling (AM). ML takes a black box approach, whose accuracy strongly depends on the representativeness of the dataset used during the initial training phase. Specifically, it can achieve very good accuracy in areas of the features' space that have been sufficiently explored during the training process. Conversely, AM techniques require no or minimal training, hence exhibiting the potential for supporting prompt instantiation of the performance model of the target system. However, in order to ensure their tractability, they typically rely on a set of simplifying assumptions. Consequently, AM's accuracy can be seriously challenged in scenarios (e.g., workload conditions) in which such assumptions are not matched.In this paper we explore several hybrid/gray box techniques that exploit AM and ML in synergy in order to get the best of the two worlds. We evaluate the proposed techniques in case studies targeting two complex and widely adopted middleware systems: a NoSQL distributed key-value store and a Total Order Broadcast (TOB) service.

This paper presents PROMPT, a PeRfOrmance Model for Partially replicated in-memory Transactional cloud stores. PROMPT combines white box Analytical Modelling and Machine Learning techniques, with the goal of achieving the best of the two methodologies: low training times, high extrapolation power, and portability across heterogeneous cloud infrastructures. We validate PROMPT via an extensive experimental study based on a popular open-source transactional in-memory data store (Red Hat's Infinispan), industry-standard benchmarks, and deployments on both public and private cloud infrastructures.

In this paper we focus on the problem of self-tuning distributed transactional cloud data stores by presenting an overview of the autonomic mechanisms integrated in the Cloud-TM platform, a transactional cloud data store developed in the context of a recent European project. Cloud-TM takes a holistic approach to self-tuning and elastic scaling, treating them as strongly intertwined problems with the ultimate goals of i) achieving optimal efficiency at any scale of the platform, and ii) minimizing resource consumption in presence of varying workloads. From a methodological perspective, this is achieved by relying on the innovative idea of exploiting the diversity of different modelling approaches, including analytical models, machine-learning and simulations. By employing these modelling techniques in synergy, the Cloud-TM platform can dynamically optimize the underlying distributed data store over a number of dimensions, including its scale, the strategy it adopts to distribute and replicate data among the platforms' nodes, as well as its replication protocol.

In this paper we investigate the issue of automatically identifying the "natural" degree of parallelism of an application using software transactional memory STM, i.e., the workload-specific multiprogramming level that maximizes application's performance. We discuss the importance of adapting the concurrency level to the workload in two different scenarios, a shared-memory and a distributed STM infrastructure. We propose and evaluate two alternative self-tuning methodologies, explicitly tailored for the considered scenarios. In shared-memory STM, we show that lightweight, black-box approaches relying solely on on-line exploration can be extremely effective. For distributed STMs, we introduce a novel hybrid approach that combines model-driven performance forecasting techniques and on-line exploration in order to take the best of the two techniques, namely enhancing robustness despite model's inaccuracies, and maximizing convergence speed towards optimum solutions.

In this paper we introduce TAS (Transactional Auto Scaler), a system for automating elastic-scaling of in-memory transactional data grids, such as NoSQL data stores or Distributed Transactional Memories. Applications of TAS range from on-line self-optimization of in-production applications to automatic generation of QoS/cost driven elastic scaling policies, and support for what-if analysis on the scalability of transactional applications.The key innovation at the core of TAS is a novel performance forecasting methodology that relies on the joint usage of analytical modeling and machine-learning. By exploiting these two, classically competing, methodologies in a synergic fashion, TAS achieves the best of the two worlds, namely high extrapolation power and good accuracy even when faced with complex workloads deployed over public cloud infrastructures.We demonstrate the accuracy and feasibility of TAS via an extensive experimental study based on a fully fledged prototype implementation, integrated with a popular open-source transactional in-memory data store (Red Hat's Infinispan), and industry-standard benchmarks generating a breadth of heterogeneous workloads.

Message batching is a well-known optimization technique to maximize throughput of networked services. The manual configuration of the appropriate batching level is however a time consuming and not trivial task. Too low batching values can in fact render the system unstable in presence of high loads, excessively high batching values, on the other hand, can lead to high latency at low load, which may be unacceptable for delay sensitive applications. The problem is further exacerbated in presence of fluctuating workloads, as in these scenarios the optimal batching level varies dynamically over time, and pursuing optimal performances demands the employment of self-adaptive mechanisms. In this paper we study the problem of self-tuning the message batching level adopting an interdisciplinary approach that employs methodologies from control theory community to optimize the performance of Total Order Broadcast (TOB), a fundamental building block to build dependable distributed systems. Specifically, we introduce an innovative self-tuning algorithm based on extremum seeking optimization principles. We provide theoretical results on its convergence properties and an extensive experimental analysis aimed at assessing the actual effectiveness of the new algorithm in a state-of-the-art group communication system.

This paper presents the design of a global memory management architecture supporting application transparent migration of simulation objects (or LPs) whose state is scattered across dynamically allocated memory chunks. Our approach is based on a non-intrusive background protocol that provides each instance of the simulation kernel with information on the current mapping of the virtual address space of all the other instances. Dynamic memory requests by the application layer are then locally mapped onto virtual-address ranges that maximize the likelihood of being portable onto the address space of a remote kernel instance. In this way, independently of the load-balancing trigger (or policy), we maximize the likelihood that a desirable migration across a specific couple of kernels can actually take place due to compliance of the corresponding source/destination address spaces. We have integrated the global memory manager within the ROme OpTimistic Simulator (ROOT-Sim), namely a run-time environment based on the optimistic synchronization paradigm which automatically and transparently parallelizes the execution of event-handler based simulation programs conforming to ANSI-C. Further, we provide a contribution in the direction of widening load balancing schemes for optimistic simulation systems by defining migration triggers and selection policies for the objects to be migrated on the basis of memory usage patterns. An experimental assessment of the architecture and of memory oriented load balancing is also provided.

We present Eagle, a new hybrid data center scheduler for data-parallel programs. Eagle dynamically divides the nodes of the data center in partitions for the execution of long and short jobs, thereby avoiding head-of-line blocking. Furthermore, it provides job awareness and avoids stragglers by a new technique, called Sticky Batch Probing (SBP). The dynamic partitioning of the data center nodes is accomplished by a technique called Succinct State Sharing (SSS), in which the distributed schedulers are informed of the locations where long jobs are executing. SSS is particularly easy to implement with a hybrid scheduler, in which the centralized scheduler places long jobs. With SBP, when a distributed scheduler places a probe for a job on a node, the probe stays there until all tasks of the job have been completed. When finishing the execution of a task corresponding to probe P, rather than executing a task corresponding to the next probe P' in its queue, the node may choose to execute another task corresponding to P. We use SBP in combination with a distributed approximation of Shortest Remaining Processing Time (SRPT) with starvation prevention. We have implemented Eagle as a Spark plugin, and we have measured job completion times for a subset of the Google trace on a 100-node cluster for a variety of cluster loads. We provide simulation results for larger clusters, different traces, and for comparison with other scheduling disciplines. We show that Eagle outperforms other state-of-the-art scheduling solutions at most percentiles, and is more robust against mis-estimation of task duration.

We prove that no fully transactional system can provide fast read transactions (including read-only ones that are considered the most frequent in practice). Specifically, to achieve fast read transactions, the system has to give up support of transactions that write more than one object. We prove this impossibility result for distributed storage systems that are causally consistent, i.e., they do not require to ensure any strong form of consistency. Therefore, our result holds also for any system that ensures a consistency level stronger than causal consistency, e.g., strict serializability. The impossibility result holds even for systems that store only two objects (and support at least two servers and at least four clients). It also holds for systems that are partially replicated. Our result justifies the design choices of state-of-the-art distributed transactional systems and insists that system designers should not put more effort to design fully-functional systems that support both fast read transactions and ensure causal or any stronger form of consistency.

This paper investigates the problem of deriving a white box performance model of Hardware Transactional Memory (HTM) systems. The proposed model targets TSX, a popular implementation of HTM integrated in Intel processors starting with the Haswell family in 2013. An inherent difficulty with building white-box models of commercially available HTM systems is that their internals are either vaguely documented or undisclosed by their manufacturers. We tackle this challenge by designing a set of experiments that allow us to shed lights on the internal mechanisms used in TSX to manage conflicts among transactions and to track their readsets and writesets. We exploit the information inferred from this experimental study to build an analytical model of TSX focused on capturing the impact on performance of two key mechanisms: the concurrency control scheme and the management of transactional meta-data in the processor's caches. We validate the proposed model by means of an extensive experimental study encompassing a broad range of workloads executed on a real system.

Transactional Causal Consistency (TCC) extends causal consistency, the strongest consistency model compatible with availability, with interactive read-write transactions, and is therefore particularly appealing for geo-replicated platforms. This paper presents Wren, the first TCC system that at the same time i) implements nonblocking read operations, thereby achieving low latency, and ii) allows an application to efficiently scale out within a replication site by sharding. Wren introduces new protocols for transaction execution, dependency tracking and stabilization. The transaction protocol supports nonblocking reads by providing a transaction with a snapshot that is the union of a fresh causal snapshot S installed by every partition in the local data center and a client-side cache for writes that are not yet included in S. The dependency tracking and stabilization protocols require only two scalar timestamps, resulting in efficient resource utilization and providing scalability in terms of replication sites. In return for these benefits, Wren slightly increases the visibility latency of updates. We evaluate Wren on an AWS deployment using up to 5 replication sites and 16 partitions per site. We show that Wren delivers up to 1.4x higher throughput and up to 3.6x lower latency when compared to the state-of-the-art design. The choice of an older snapshot increases local update visibility latency by a few milliseconds. The use of only two timestamps to track causality increases remote update visibility latency by less than 15%.

The vast majority of data center schedulers use task runtime estimates to improve the quality of their scheduling decisions. Knowledge about runtimes allows the schedulers, among other things, to achieve better load balance and to avoid head-of-line blocking. Obtaining accurate runtime estimates is, however, far from trivial, and erroneous estimates lead to sub-optimal scheduling decisions. Techniques to mitigate the effect of inaccurate estimates have shown some success, but the fundamental problem remains. This paper presents Kairos, a novel data center scheduler that assumes no prior information on task runtimes. Kairos introduces a distributed approximation of the Least Attained Service (LAS) scheduling policy. Kairos consists of a centralized scheduler and per-node schedulers. The per-node schedulers implement LAS for tasks on their node, using preemption as necessary to avoid head-of-line blocking. The centralized scheduler distributes tasks among nodes in a manner that balances the load and imposes on each node a workload in which LAS provides favorable performance. We have implemented Kairos in YARN. We compare its performance against the YARN FIFO scheduler and Big-C, an open-source state-of-the-art YARN-based scheduler that also uses preemption. Compared to YARN FIFO, Kairos reduces the median job completion time by 73% and the 99th percentile by 30%. Compared to Big-C, the improvements are 37% for the median and 57% for the 99th percentile. We evaluate Kairos at scale by implementing it in the Eagle simulator and comparing its performance against Eagle. Kairos improves the 99th percentile of short job completion times by up to 55% for the Google trace and 85% for the Yahoo trace.

Causal consistency is an attractive consistency model for geo-replicated data stores. It is provably the strongest model that tolerates network partitions. It avoids the long latencies associated with strong consistency, and, especially when using read-only transactions (ROTs), it prevents many of the anomalies of weaker consistency models. Recent work has shown that causal consistency allows "latency-optimal" ROTs, that are nonblocking, single-round and single-version in terms of communication. On the surface, this latency optimality is very appealing, as the vast majority of applications are assumed to have read-dominated workloads., In this paper, we show that such "latency-optimal" ROTs induce an extra overhead on writes that is so high that it actually jeopardizes performance even in read-dominated workloads. We show this result from a practical as well as from a theoretical angle., We present the Contrarian protocol that implements "almost latency-optimal" ROTs, but that does not impose on the writes any of the overheads incurred by latency-optimal protocols. In Contrarian, ROTs are nonblocking and single-version, but they require two rounds of client-server communication. We experimentally show that this protocol not only achieves higher throughput, but, surprisingly, also provides better latencies for all but the lowest loads and the most read-heavy workloads., We furthermore prove that the extra overhead imposed on writes by latency-optimal ROTs is inherent, i.e., it is not an artifact of the design we consider, and cannot be avoided by any implementation of latency-optimal ROTs. We show in particular that this overhead grows linearly with the number of clients.

Causal consistency is an attractive consistency model for geo-replicated data stores because it hits a sweet spot in the ease of programmability vs performance trade-off. In this paper we propose a new approach to causal consistency, which we call Optimistic Causal Consistency (OCC). The optimism of our approach lies in the fact that updates from a remote data center are immediately made visible to clients in the local data center. A client, hence, always reads the freshest version of an item, whose dependencies, however, might have not been installed in the local data center yet. When serving a read request, a server can detect whether it has not received such dependencies yet. This is achieved without inter-server synchronization thanks to cheap dependency meta-data supplied by the client. Upon detecting a missing dependency, the server waits to receive it. This approach contrasts with the design of existing systems, which are prone to expose stale versions of a data items, to ensure that clients only see versions whose dependencies have already been replicated in the local data center. OCC explores a novel trade-off in the landscape of consistency models. Because network partitions are practically rare events, OCC partially trades availability to improve other performance metrics. On the one side, OCC maximizes the freshness of data returned to clients and reduces the communication overhead. On the other side, a server might need to wait before serving a client’s request, leading the system to be unavailable in case of a network partition. To overcome this limitation, we propose a recovery mechanism that allows an OCC system to fall back to a pessimistic protocol to recover availability. We implement OCC in a new system, which we call POCC. We compare POCC against a recent (pessimistic) approach to causal consistency using heterogeneous workloads on an Amazon AWS deployment encompassing up to 96 nodes scattered over 3 data centers. We show that POCC is able to maximize the freshness of data returned to client while providing comparable or better performance than its pessimistic counterpart in a wide range of production-like workloads.

Many different mechanisms have been developed to implement Distributed Transactional Memory (DTM). Unfortunately, there is no “one-size-fits-all” design that offers the desirable performance across all possible workloads and scales. In fact, the performance of these mechanisms is affected by a number of intertwined factors that make it hard, or even impossible, to statically configure a DTM platform for optimal performance. These observations have motivated the emergence of self-tuning schemes for automatically adapting the algorithms and parameters used by the main building blocks of DTM systems. This chapter surveys existing research in the area of autonomic DTM design, with a focus on the approaches aimed at answering the following two fundamental questions: how many resources (number of nodes, etc.) should a DTM platform be provisioned with, and which protocols should be used to ensure data consistency.

We prove that transactions cannot be fast in an asynchronous fault-tolerant system. Our result holds in any system where we require transactions to ensure monotonic writes, or any stronger consistency model, such as, causal consistency. Thus, our result unveils an important, and so far unknown, limitation of fast transactions: they are impossible if we want to tolerate the failure of even one server.