In this report, we consider the impact of the consistency model on checkpointing and rollback algorithms for distributed shared memory. In particular, we consider specific implementations of four consistency models for distributed shared memory, namely, linearizability, sequential consistency, causal consistency and eventual consistency, and develop checkpointing and rollback algorithms that can be integrated into the implementations of the consistency models. Our results empirically demonstrate that the mechanisms used to implement stronger consistency models lead to simpler or more efficient checkpointing algorithms.
We present the checkpointing scheme of Abacus, an $N$-body simulation code that allocates all persistent state in POSIX shared memory, or ramdisk. Checkpointing becomes as simple as copying files from ramdisk to external storage. The main simulation executable is invoked once per time step, memory mapping the input state, computing the output state directly into ramdisk, and unmapping the input state. The main executable remains unaware of the concept of checkpointing, with the top-level driver code launching a file-system copy between executable invocations when a checkpoint is needed. Since …
Unified Virtual Memory (UVM) was recently introduced on recent NVIDIA GPUs. Through software and hardware support, UVM provides a coherent shared memory across the entire heterogeneous node, migrating data as appropriate. The older CUDA programming style is akin to older large-memory UNIX applications which used to directly load and unload memory segments. Newer CUDA programs have started taking advantage of UVM for the same reasons of superior programmability that UNIX applications long ago switched to assuming the presence of virtual memory. Therefore, checkpointing of UVM will become increa…
The focus of this paper is on causal consistency in a {\em partially replicated} distributed shared memory (DSM) system that provides the abstraction of shared read/write registers. Maintaining causal consistency in distributed shared memory systems has received significant attention in the past, mostly on {\em full replication} wherein each replica stores a copy of all the registers in the shared memory. To ensure causal consistency, all causally preceding updates must be performed before an update is performed at any given replica. Therefore, some mechanism for tracking causal dependencies i…
In distributed systems where strong consistency is costly when not impossible, causal consistency provides a valuable abstraction to represent program executions as partial orders. In addition to the sequential program order of each computing entity, causal order also contains the semantic links between the events that affect the shared objects -- messages emission and reception in a communication channel , reads and writes on a shared register. Usual approaches based on semantic links are very difficult to adapt to other data types such as queues or counters because they require a specific an…
Concurrent linearizable access to shared objects can be prohibitively expensive in a high contention workload. Many applications apply ad-hoc techniques to eliminate the need of synchronous atomic updates, which may result in non-linearizable implementations. We propose a new programming model which leverages such patterns for concurrent access to objects in a shared memory system. In this model, each thread maintains different views on the shared object - a thread-local view and a global view. As the thread-local view is not shared, it can be updated without incurring synchronization costs. T…
This work examines strategies to handle large shared data objects in distributed storage systems (DSS), while boosting the number of concurrent accesses, maintaining strong consistency guarantees, and ensuring good operation performance. To this respect, we define the notion of fragmented objects:con-current objects composed of a list of fragments (or blocks) that allow operations to manipulate each of their fragments individually. As the fragments belong to the same object, it is not enough that each fragment is linearizable to have useful consistency guarantees in the composed object. Hence,…
In distributed applications, Brewer's CAP theorem tells us that when networks become partitioned (P), one must give up either consistency (C) or availability (A). Consistency is agreement on the values of shared variables; availability is the ability to respond to reads and writes accessing those shared variables. Availability is a real-time property whereas consistency is a logical property. We have extended the CAP theorem to relate quantitative measures of these two properties to quantitative measures of communication and computation latency (L), obtaining a relation called the CAL theorem …
Deduplication has been largely employed in distributed storage systems to improve space efficiency. Traditional deduplication research ignores the design specifications of shared-nothing distributed storage systems such as no central metadata bottleneck, scalability, and storage rebalancing. Further, deduplication introduces transactional changes, which are prone to errors in the event of a system failure, resulting in inconsistencies in data and deduplication metadata. In this paper, we propose a robust, fault-tolerant and scalable cluster-wide deduplication that can eliminate duplicate copie…
To implement a linearizable shared memory in synchronous message-passing systems it is necessary to wait for a time linear to the uncertainty in the latency of the network for both read and write operations. Waiting only for one of them suffices for sequential consistency. This paper extends this result to crash-prone asynchronous systems, proposing a distributed algorithm building a sequentially consistent shared snapshot memory on top of an asynchronous message-passing system where less than half of the processes may crash. We prove that waiting is needed only when a process invokes a read/s…
It has been proved that to implement a linearizable shared memory in synchronous message-passing systems it is necessary to wait for a time proportional to the uncertainty in the latency of the network for both read and write operations, while waiting during read or during write operations is sufficient for sequential consistency. This paper extends this result to crash-prone asynchronous systems. We propose a distributed algorithm that builds a sequentially consistent shared memory abstraction with snapshot on top of an asynchronous message-passing system where less than half of the processes…
Shared Memory is a mechanism that allows several processes to communicate with each other by accessing -- writing or reading -- a set of variables that they have in common. A Consistency Model defines how each process observes the state of the Memory, according to the accesses performed by it and by the rest of the processes in the system. Therefore, it determines what value a read returns when a given process issues it. This implies that there must be an agreement among all, or among processes in different subsets, on the order in which all or a subset of the accesses happened. It is clear th…
A new memory coherence protocol, Tardis, is proposed. Tardis uses timestamp counters representing logical time as well as physical time to order memory operations and enforce sequential consistency in any type of shared memory system. Tardis is unique in that as compared to the widely-adopted directory coherence protocol, and its variants, it completely avoids multicasting and only requires O(log N) storage per cache block for an N-core system rather than O(N) sharer information. Tardis is simpler and easier to reason about, yet achieves similar performance to directory protocols on a wide ran…
The behavioural theory of concurrent systems states that any concurrent system can be captured by a behaviourally equivalent concurrent Abstract State Machine (cASM). While the theory in general assumes shared locations, it remains valid, if different agents can only interact via messages, i.e. sharing is restricted to mailboxes. There may even be a strict separation between memory managing agents and other agents that can only access the shared memory by sending query and update requests to the memory agents. This article is dedicated to an investigation of replicated data that is maintained …
Most work on the verification of concurrent objects for shared memory assumes sequential consistency, but most multicore processors support only weak memory models that do not provide sequential consistency. Furthermore, most verification efforts focus on the linearizability of concurrent objects, but there are existing implementations optimized to run on weak memory models that are not linearizable. In this paper, we address these problems by introducing causal linearizability, a correctness condition for concurrent objects running on weak memory models. Like linearizability itself, causal li…
In distributed applications, Brewer's CAP theorem tells us that when networks become partitioned, there is a tradeoff between consistency and availability. Consistency is agreement on the values of shared variables across a system, and availability is the ability to respond to reads and writes accessing those shared variables. We quantify these concepts, giving numerical values to inconsistency and unavailability. Recognizing that network partitioning is not an all-or-nothing proposition, we replace the P in CAP with L, a numerical measure of apparent latency, and derive the CAL theorem, an al…
We present the SC-ABD algorithm that implements sequentially consistent distributed shared memory (DSM). The algorithm tolerates that less than half of the processes are faulty (crash-stop). Compared to the multi-writer ABD algorithm, SC-ABD requires one instead of two round-trips of communication to perform a write operation, and an equal number of round-trips (two) to perform a read operation. Although sequential consistency is not a compositional consistency condition, the provided correctness proof is compositional.
Data replication is used in distributed systems to maintain up-to-date copies of shared data across multiple computers in a network. However, despite decades of research, algorithms for achieving consistency in replicated systems are still poorly understood. Indeed, many published algorithms have later been shown to be incorrect, even some that were accompanied by supposed mechanised proofs of correctness. In this work, we focus on the correctness of Conflict-free Replicated Data Types (CRDTs), a class of algorithm that provides strong eventual consistency guarantees for replicated data. We de…
Memory consistency models (MCMs) specify the legal ordering and visibility of shared memory accesses in a parallel program. Traditionally, instruction set architecture (ISA) MCMs assume that relevant program-visible memory ordering behaviors only result from shared memory interactions that take place between user-level program instructions. This assumption fails to account for virtual memory (VM) implementations that may result in additional shared memory interactions between user-level program instructions and both 1) system-level operations (e.g., address remappings and translation lookaside…
