Browsing by Author "Palmieri, Roberto"

Now showing 1 - 9 of 9
  • Thumbnail Image
    Extracting Parallelism from Legacy Sequential Code Using Transactional Memory
    Saad Ibrahim, Mohamed Mohamed (Virginia Tech, 2016-07-26)
    Increasing the number of processors has become the mainstream for the modern chip design approaches. However, most applications are designed or written for single core processors; so they do not benefit from the numerous underlying computation resources. Moreover, there exists a large base of legacy software which requires an immense effort and cost of rewriting and re-engineering to be made parallel. In the past decades, there has been a growing interest in automatic parallelization. This is to relieve programmers from the painful and error-prone manual parallelization process, and to cope with new architecture trend of multi-core and many-core CPUs. Automatic parallelization techniques vary in properties such as: the level of paraellism (e.g., instructions, loops, traces, tasks); the need for custom hardware support; using optimistic execution or relying on conservative decisions; online, offline or both; and the level of source code exposure. Transactional Memory (TM) has emerged as a powerful concurrency control abstraction. TM simplifies parallel programming to the level of coarse-grained locking while achieving fine-grained locking performance. This dissertation exploits TM as an optimistic execution approach for transforming a sequential application into parallel. The design and the implementation of two frameworks that support automatic parallelization: Lerna and HydraVM, are proposed, along with a number of algorithmic optimizations to make the parallelization effective. HydraVM is a virtual machine that automatically extracts parallelism from legacy sequential code (at the bytecode level) through a set of techniques including code profiling, data dependency analysis, and execution analysis. HydraVM is built by extending the Jikes RVM and modifying its baseline compiler. Correctness of the program is preserved through exploiting Software Transactional Memory (STM) to manage concurrent and out-of-order memory accesses. Our experiments show that HydraVM achieves speedup between 2×-5× on a set of benchmark applications. Lerna is a compiler framework that automatically and transparently detects and extracts parallelism from sequential code through a set of techniques including code profiling, instrumentation, and adaptive execution. Lerna is cross-platform and independent of the programming language. The parallel execution exploits memory transactions to manage concurrent and out-of-order memory accesses. This scheme makes Lerna very effective for sequential applications with data sharing. This thesis introduces the general conditions for embedding any transactional memory algorithm into Lerna. In addition, the ordered version of four state-of-art algorithms have been integrated and evaluated using multiple benchmarks including RSTM micro benchmarks, STAMP and PARSEC. Lerna showed great results with average 2.7× (and up to 18×) speedup over the original (sequential) code. While prior research shows that transactions must commit in order to preserve program semantics, placing the ordering enforces scalability constraints at large number of cores. In this dissertation, we eliminates the need for commit transactions sequentially without affecting program consistency. This is achieved by building a cooperation mechanism in which transactions can forward some changes safely. This approach eliminates some of the false conflicts and increases the concurrency level of the parallel application. This thesis proposes a set of commit order algorithms that follow the aforementioned approach. Interestingly, using the proposed commit-order algorithms the peak gain over the sequential non-instrumented execution in RSTM micro benchmarks is 10× and 16.5× in STAMP. Another main contribution is to enhance the concurrency and the performance of TM in general, and its usage for parallelization in particular, by extending TM primitives. The extended TM primitives extracts the embedded low level application semantics without affecting TM abstraction. Furthermore, as the proposed extensions capture common code patterns, it is possible to be handled automatically through the compilation process. In this work, that was done through modifying the GCC compiler to support our TM extensions. Results showed speedups of up to 4× on different applications including micro benchmarks and STAMP. Our final contribution is supporting the commit-order through Hardware Transactional Memory (HTM). HTM contention manager cannot be modified because it is implemented inside the hardware. Given such constraint, we exploit HTM to reduce the transactional execution overhead by proposing two novel commit order algorithms, and a hybrid reduced hardware algorithm. The use of HTM improves the performance by up to 20% speedup.
  • Thumbnail Image
    High Performance Inter-kernel Communication and Networking in a Replicated-kernel Operating System
    Ansary, B M Saif (Virginia Tech, 2016-01-20)
    Modern computer hardware platforms are moving towards high core-count and heterogeneous Instruction Set Architecture (ISA) processors to achieve improved performance as single core performance has reached its performance limit. These trends put the current monolithic SMP operating system (OS) under scrutiny in terms of scalability and portability. Proper pairing of computing workloads with computing resources has become increasingly arduous with traditional software architecture. One of the most promising emerging operating system architectures is the Multi-kernel. Multi-kernels not only address scalability issues, but also inherently support heterogeneity. Furthermore, provide an easy way to properly map computing workloads to the correct type of processing resources in presence of heterogeneity. Multi-kernels do so by partitioning the resources and running independent kernel instances and co-operating amongst themselves to present a unified view of the system to the application. Popcorn is one the most prominent multi-kernels today, which is unique in the sense that it runs multiple Linux instances on different cores or group of cores, and provides a unified view of the system i.e., Single System Image (SSI). This thesis presents four contributions. First, it introduces a filesystem for Popcorn, which is a vital part to provide a SSI. Popcorn supports thread/process migration that requires migration of file descriptors which is not provided by traditional filesystems as well as popular distributed file systems, this work proposes a scalable messaging based file descriptor migration and consistency protocol for Popcorn. Second, multi-kernel OSs rely heavily on a fast low latency messaging layer to be scalable. Messaging is even more important in heterogeneous systems where different types of cores are on different islands with no shared memory. Thus, another contribution proposes a fast-low latency messaging layer to enable communication among heterogeneous processor islands for Heterogeneous Popcorn. With advances in networking technology, newest Ethernet technologies are able to support up to 40 Gbps bandwidth, but due to scalability issues in monolithic kernels, the number of connections served per second does not scale with this increase in speed.Therefore, the third and fourth contributions try to address this problem with Snap Bean, a virtual network device and Angel, an opportunistic load balancer for Popcorn's network system. With the messaging layer Popcorn gets over 30% performance benefit over OpenCL and Intel Offloading technique (LEO). And with NetPopcorn we achieve over 7 to 8 times better performance over vanilla Linux and 2 to 5 times over state-of-the-art Affinity Accept.
  • Thumbnail Image
    Improving Performance of Highly-Programmable Concurrent Applications by Leveraging Parallel Nesting and Weaker Isolation Levels
    Niles, Duane Francis Jr. (Virginia Tech, 2015-07-15)
    The recent development of multi-core computer architectures has largely affected the creation of everyday applications, requiring the adoption of concurrent programming to significantly utilize the divided processing power of computers. Applications must be split into sections able to execute in parallel, without any of these sections conflicting with one another, thereby necessitating some form of synchronization to be declared. The most commonly used methodology is lock-based synchronization; although, to improve performance the most, developers must typically form complex, low-level implementations for large applications, which can easily create potential errors or hindrances. An abstraction from database systems, known as transactions, is a rising concurrency control design aimed to circumvent the challenges with programmability, composability, and scalability in lock-based synchronization. Transactions execute their operations speculatively and are capable of being restarted (or rolled back) when there exist conflicts between concurrent actions. As such issues can occur later in the lifespans of transactions, entire rollbacks are not that effective for performance. One particular method, known as nesting, was created to counter that drawback. Nesting is the act of enclosing transactions within other transactions, essentially dividing the work into pieces called sub-transactions. These sub-transactions can roll back without affecting the entire main transaction, although general nesting models only allow one sub-transaction to perform work at a time. The first main contribution in this thesis is SPCN, an algorithm that parallelizes nested transactions while automatically processing any potential conflicts that may arise, eliminating the burden of additional processing from the application developers. Two versions of SPCN exist: Strict, which enforces the sub-transactions' work to be made visible in a serialized order; and Relaxed, which allows sub-transactions to distribute their information immediately as they finish (therefore invalidation may occur after-the-fact and must be handled). Despite the additional logic required by SPCN, it outperforms traditional closed nesting by 1.78x at the lowest and 3.78x at the highest in the experiments run. Another method to alter transactional execution and boost performance is to relax the rules of visibility for parallel operations (known as their isolation). Depending on the application, correctness is not broken even if some transactions see external work that may later be undone due to a rollback, or if an object is written while another transaction is using an older instance of its data. With lock-based synchronization, developers would have to explicitly design their application with varying amounts of locks, and different lock organizations or hierarchies, to change the strictness of the execution. With transactional systems, the processing performed by the system itself can be set to utilize different rulings, which can change the performance of an application without requiring it to be largely redesigned. This notion leads to the second contribution in this thesis: AsR, or As-Serializable transactions. Serializability is the general form of isolation or strictness for transactions in many applications. In terms of execution, its definition is equivalent to only one transaction running at a time in a given system. Many transactional systems use their own internal form of locking to create Serializable executions, but it is typically too strict for many applications. AsR transactions allow the internal processing to be relaxed while additional meta-data is used external to the system, without requiring any interaction from the developer or any changes to the given application. AsR transactions offer multiple orders of magnitude more in throughput in highly-contentious scenarios, due to their capability to outlast traditional levels of isolation.
  • Thumbnail Image
    Mutex Locking versus Hardware Transactional Memory: An Experimental Evaluation
    Moore, Sean Ryan (Virginia Tech, 2015-09-16)
    It has historically been the case that CPUs have run programs ever faster without significant intervention on the behalf of the programmer. However, this "free lunch" has largely ended due to the end of exponentially increasing core frequency and the current slow increase in instruction-level parallelism but continues to a degree in cache size improvements. But since Moore's law still largely continues "lunch", i.e. program performance, can still be bought at the price of rewriting code for multiple cores, which is enabled by the trend Moore's law describes. Multicore architectures cannot aid performance for problems whose solutions are necessarily sequential in nature and writing efficient and correct concurrent programs is not easy in all cases when using synchronization methods like fine-grained mutex locks. Transactional memory, and its implementation as hardware transactional memory, allow programmers to write concurrent applications without the attendant complexity of programming with mutex locks. This allows programmers to focus on optimizing the application for performance. Given that transactions can run two segments of code in parallel that a mutex lock would force to run sequentially and that transactions can abort, causing a program to do the same work more than once, whether transactions perform better or worse than mutex locks is dependent on the program's execution profile and the coarseness or fineness at which mutex locks are used. In this thesis the GNU C Library's futex implementation of mutex locks and Intel's Restricted Transactional Memory have been compared and the behavior of those transactions has been analyzed. This analysis includes a pathological behavior permitted by the GNU C Library's hardware transactional memory implementation of mutex locks. The tradeoffs between fine-grained and global locking implementations have been discussed, compared, and used in the context of fallback locks for hardware transactions. This thesis provides evidence to the effect that fine-grained locking is not critical for program performance and that in many cases global locking and hardware transactions can provide nearly equivalent performance without the programming difficulties. This work has shown that across the 23 applications examined, with relation to their original locking implementation, a global locking scheme without elision has a 0.96x speedup, Intel's Restricted Transactional Memory (RTM) with the application's original locks as a fallback has a 1.01x speedup and with global lock fallback RTM has a speedup of 0.97x. This work is supported in part by NAVSEA/NEEC under grant 3003279297. Any opinions, findings, and conclusions or recommendations expressed in this thesis are those of the author and do not necessarily reflect the views of NAVSEA.
  • Thumbnail Image
    On Improving Distributed Transactional Memory through Nesting, Partitioning and Ordering
    Turcu, Alexandru (Virginia Tech, 2015-03-03)
    Distributed Transactional Memory (DTM) is an emerging, alternative concurrency control model that aims to overcome the challenges of distributed-lock based synchronization. DTM employs transactions in order to guarantee consistency in a concurrent execution. When two or more transactions conflict, all but one need to be delayed or rolled back. Transactional Memory supports code composability by nesting transactions. Nesting how- ever can be used as a strategy to improve performance. The closed nesting model enables partial rollback by allowing a sub-transaction to abort without aborting its parent, thus reducing the amount of work that needs to be retried. In the open nesting model, sub- transactions can commit to the shared state independently of their parents. This reduces isolation and increases concurrency. Our first main contribution in this dissertation are two extensions to the existing Transac- tional Forwarding Algorithm (TFA). Our extensions are N-TFA and TFA-ON, and support closed nesting and open nesting, respectively. We additionally extend the existing SCORe algorithm with support for open nesting (we call the result SCORe-ON). We implement these algorithms in a Java DTM framework and evaluate them. This represents the first study of transaction nesting in the context of DTM, and contributes the first DTM implementation which supports closed nesting or open nesting. Closed nesting through our N-TFA implementation proved insufficient for any significant throughput improvements. It ran on average 2% faster than flat nesting, while performance for individual tests varied between 42% slowdown and 84% speedup. The workloads that benefit most from closed nesting are characterized by short transactions, with between two and five sub-transactions. Open nesting, as exemplified by our TFA-ON and SCORe-ON implementations, showed promising results. We determined performance improvement to be a trade-off between the overhead of additional commits and the fundamental conflict rate. For write-intensive, high- conflict workloads, open nesting may not be appropriate, and we observed a maximum speedup of 30%. On the other hand, for lower fundamental-conflict workloads, open nesting enabled speedups of up to 167% in our tests. In addition to the two nesting algorithms, we also develop Hyflow2, a high-performance DTM framework for the Java Virtual Machine, written in Scala. It has a clean Scala API and a compatibility Java API. Hyflow2 was on average two times faster than Hyflow on high-contention workloads, and up to 16 times faster in low-contention workloads. Our second main contribution for improving DTM performance is automated data partition- ing. Modern transactional processing systems need to be fast and scalable, but this means many such systems settled for weak consistency models. It is however possible to achieve all of strong consistency, high scalability and high performance, by using fine-grained partitions and light-weight concurrency control that avoids superfluous synchronization and other over- heads such as lock management. Independent transactions are one such mechanism, that rely on good partitions and appropriately defined transactions. On the downside, it is not usually straightforward to determine optimal partitioning schemes, especially when dealing with non-trivial amounts of data. Our work attempts to solve this problem by automating the partitioning process, choosing the correct transactional primitive, and routing transactions appropriately. Our third main contribution is Alvin, a system for managing concurrently running trans- actions on a geographically replicated data-store. Alvin supports general-purpose transactions, and guarantees strong consistency criteria. Through a novel partial order broadcast protocol, Alvin maximizes the parallelism of ordering and local transaction processing, resulting in low client-perceived latency. Alvin can process read-only transactions either lo- cally or globally, according to the desired consistency criterion. Conflicting transactions are ordered across all sites. We built Alvin in the Go programming language. We conducted our evaluation study on Amazon EC2 infrastructure and compared against Paxos- and EPaxos- based state machine replication protocols. Our results reveal that Alvin provides significant speed-up for read-dominated TPC-C workloads: as much as 4.8x when compared to EPaxos on 7 datacenters, and up to 26% in write-intensive workloads. Our fourth and final contribution is M2Paxos, a multi-leader implementation of Generalized Consensus. Single leader-based consensus protocols are known to stop scaling once the leader reaches its saturation point. Ordering commands based on conflicts is appealing due to the potentially higher parallelism, but is imperfect due to the higher quorum sizes required for fast decisions and the need to compare commands and track their dependencies. M2Paxos on the other hand exploits fast decisions (i.e., delivery of a command in two communication delays) by leveraging a classic quorum size, matching a majority of nodes deployed. M2Paxos does not establish command dependencies based on conflicts, but it binds accessed objects to nodes, making sure commands operating on the same object will be ordered by the same node. Our evaluation study of M2Paxos (also built in Go) confirms the effectiveness of this approach, getting up to 7⨉ improvements in performance over state- of-the-art consensus and generalized consensus algorithms.
  • Thumbnail Image
    On Optimizing Transactional Memory: Transaction Splitting, Scheduling, Fine-grained Fallback, and NUMA Optimization
    Mohamedin, Mohamed Ahmed Mahmoud (Virginia Tech, 2015-09-01)
    The industrial shift from single core processors to multi-core ones introduced many challenges. Among them, a program cannot get a free performance boost by just upgrading to a new hardware because new chips include more processing units but at the same (or comparable) clock speed as the previous generation. In order to effectively exploit the new available hardware and thus gain performance, a program should maximize parallelism. Unfortunately, parallel programming poses several challenges, especially when synchronization is involved because parallel threads need to access the same shared data. Locks are the standard synchronization mechanism but gaining performance using locks is difficult for a non-expert programmers and without deeply knowing the application logic. A new, easier, synchronization abstraction is therefore required and Transactional Memory (TM) is the concrete candidate. TM is a new programming paradigm that simplifies the implementation of synchronization. The programmer just defines atomic parts of the code and the underlying TM system handles the required synchronization, optimistically. In the past decade, TM researchers worked extensively to improve TM-based systems. Most of the work has been dedicated to Software TM (or STM) as it does not requires special transactional hardware supports. Very recently (in the past two years), those hardware supports have become commercially available as commodity processors, thus a large number of customers can finally take advantage of them. Hardware TM (or HTM) provides the potential to obtain the best performance of any TM-based systems, but current HTM systems are best-effort, thus transactions are not guaranteed to commit in any case. In fact, HTM transactions are limited in size and time as well as prone to livelock at high contention levels. Another challenge posed by the current multi-core hardware platforms is their internal architecture used for interfacing with the main memory. Specifically, when the common computer deployment changed from having a single processor to having multiple multi-core processors, the architects redesigned also the hardware subsystem that manages the memory access from the one providing a Uniform Memory Access (UMA), where the latency needed to fetch a memory location is the same independently from the specific core where the thread executes on, to the current one with a Non-Uniform Memory Access (NUMA), where such a latency differs according to the core used and the memory socket accessed. This switch in technology has an implication on the performance of concurrent applications. In fact, the building blocks commonly used for designing concurrent algorithms under the assumptions of UMA (e.g., relying on centralized meta-data) may not provide the same high performance and scalability when deployed on NUMA-based architectures. In this dissertation, we tackle the performance and scalability challenges of multi-core architectures by providing three solutions for increasing performance using HTM (i.e., Part-HTM, Octonauts, and Precise-TM), and one solution for solving the scalability issues provided by NUMA-architectures (i.e., Nemo). • Part-HTM is the first hybrid transactional memory protocol that solves the problem of transactions aborted due to the resource limitations (space/time) of current best-effort HTM. The basic idea of Part-HTM is to partition those transactions into multiple sub-transactions, which can likely be committed in hardware. Due to the eager nature of HTM, we designed a low-overhead software framework to preserve transaction's correctness (with and without opacity) and isolation. Part-HTM is efficient: our evaluation study confirms that its performance is the best in all tested cases, except for those where HTM cannot be outperformed. However, in such a workload, Part-HTM still performs better than all other software and hybrid competitors. • Octonauts tackles the live-lock problem of HTM at high contention level. HTM lacks of advanced contention management (CM) policies. Octonauts is an HTM-aware scheduler that orchestrates conflicting transactions. It uses a priori knowledge of transactions' working-set to prevent the activation of conflicting transactions, simultaneously. Octonauts also accommodates both HTM and STM with minimal overhead by exploiting adaptivity. Based on the transaction's size, time, and irrevocable calls (e.g., system call) Octonauts selects the best path among HTM, STM, or global locking. Results show a performance improvement up to 60% when Octonauts is deployed in comparison with pure HTM with falling back to global locking. • Precise-TM is a unique approach to solve the granularity of the software fallback path of best-efforts HTM. It provide an efficient and precise technique for HTM-STM communication such that HTM is not interfered by concurrent STM transactions. In addition, the added overhead is marginal in terms of space or execution time. Precise-TM uses address-embedded locks (pointers bit-stealing) for a precise communication between STM and HTM. Results show that our precise fine-grained locking pays off as it allows more concurrency between hardware and software transactions. Specifically, it gains up to 5x over the default HTM implementation with a single global lock as fallback path. • Nemo is a new STM algorithm that ensures high and scalable performance when an application workload with a data locality property is deployed. Existing STM algorithms rely on centralized shared meta-data (e.g., a global timestamp) to synchronize concurrent accesses, but in such a workload, this scheme may hamper the achievement of scalable performance given the high latency introduced by NUMA architectures for updating those centralized meta-data. Nemo overcomes these limitations by allowing only those transactions that actually conflict with each other to perform inter-socket communication. As a result, if two transactions are non-conflicting, they cannot interact with each other through any meta-data. Such a policy does not apply for application threads running in the same socket. In fact, they are allowed to share any meta-data even if they execute non-conflicting operations because, supported by our evaluation study, we found that the local processing happening inside one socket does not interfere with the work done by parallel threads executing on other sockets. Nemo's evaluation study shows improvement over state-of-the-art TM algorithms by as much as 65%.
  • Thumbnail Image
    On the Fault-tolerance and High Performance of Replicated Transactional Systems
    Hirve, Sachin (Virginia Tech, 2015-09-28)
    With the recent technological developments in last few decades, there is a notable shift in the way business/consumer transactions are conducted. These transactions are usually triggered over the internet and transactional systems working in the background ensure that these transactions are processed. The majority of these transactions nowadays fall in Online Transaction Processing (OLTP) category, where low latency is preferred characteristic. In addition to low latency, OLTP transaction systems also require high service continuity and dependability. Replication is a common technique that makes the services dependable and therefore helps in providing reliability, availability and fault-tolerance. Deferred Update Replication (DUR) and Deferred Execution Replication (DER) represent the two well known transaction execution models for replicated transactional systems. Under DUR, a transaction is executed locally at one node before a global certification is invoked to resolve conflicts against other transactions running on remote nodes. On the other hand, DER postpones the transaction execution until the agreement on a common order of transaction requests is reached. Both DUR and DER require a distributed ordering layer, which ensures a total order of transactions even in case of faults. In today's distributed transactional systems, performance is of paramount importance. Any loss in performance, e.g., increased latency due to slow processing of client requests, may entail loss of revenue for businesses. On one hand, the DUR model is a good candidate for transaction processing in those systems in case the conflicts among transactions are rare, while it can be detrimental for high conflict workload profiles. On the other hand, the DER model is an attractive choice because of its ability to behave as independent of the characteristics of the workload, but trivial realizations of the model ultimately do not offer a good performance increase margin. Indeed transactions are executed sequentially and the total order layer can be a serious bottleneck for latency and scalability. This dissertation proposes novel solutions and system optimizations to enhance the overall performance of replicated transactional systems. The first presented result is HiperTM, a DER-based transaction replication solution that is able to alleviate the costs of the total order layer via speculative execution techniques. HiperTM exploits the time that is between the broadcast of a client request and the finalization of the order for that request to speculatively execute the request, so to achieve an overlapping between replicas coordination and transactions execution. HiperTM proposes two main components: OS-Paxos, a novel total order layer that is able to early deliver requests optimistically according to a tentative order, which is then either confirmed or rejected by a final total order; SCC, a lightweight speculative concurrency control protocol that is able to exploit the optimistic delivery of OS-Paxos and execute transactions in a speculative fashion. SCC still processes write transactions serially in order to minimize the code instrumentation overheads, but it is able to parallelize the execution of read-only transactions thanks to its built-in object multiversion scheme. The second contribution in this dissertation is X-DUR, a novel transaction replication system that addressed the high cost of local and remote aborts in case of high contention on shared objects in DUR based approaches, due to which the performance is adversely affected. Exploiting the knowledge of client's transaction locality, X-DUR incorporates the benefits of state machine approach to scale-up the distributed performance of DUR systems. As third contribution, this dissertation proposes Archie, a DER-based replicated transactional system that improves HiperTM in two aspects. First, Archie includes a highly optimized total order layer that combines optimistic-delivery and batching thus allowing the anticipation of a big amount of work before the total order is finalized. Then the concurrency control is able to process transactions speculatively and with a higher degree of parallelism, although the order of the speculative commits still follows the order defined by the optimistic delivery. Both HiperTM and Archie perform well up to a certain number of nodes in the system, beyond which their performance is impacted by limitations of single leader-based total-order layer. This motivates the design of Caesar, the forth contribution of this dissertation, which is a transactional system based on a novel multi-leader partial order protocol. Caesar enforces a partial order on the execution of transactions according to their conflicts, by letting non-conflicting transactions to proceed in parallel and without enforcing any synchronization during the execution (e.g., no locks). As the last contribution, this dissertation presents Dexter, a replication framework that exploits the commonly observed phenomenon such that not all read-only workloads require up-to-date data. It harnesses the application specific freshness and content-based constraints of read-only transactions to achieve high scalability. Dexter services the read-only requests according to the freshness guarantees specified by the application and routes the read-only workload accordingly in the system to achieve high performance and low latency. As a result, Dexter framework also alleviates the interference between read-only requests and read-write requests thereby helping to improve the performance of read-write requests execution as well.
  • Thumbnail Image
    Optimizing Distributed Transactions: Speculative Client Execution, Certified Serializability, and High Performance Run-Time
    Pandey, Utkarsh (Virginia Tech, 2016-09-01)
    On-line services already form an important part of modern life with an immense potential for growth. Most of these services are supported by transactional systems, which are backed by database management systems (DBMS) in many cases. Many on-line services use replication to ensure high-availability, fault tolerance and scalability. Replicated systems typically consist of different nodes running the service co-ordinated by a distributed algorithm which aims to drive all the nodes along the same sequence of states by providing a total order to their operations. Thus optimization of both local DBMS operations through concurrency control and the distributed algorithm driving replicated services can lead to enhancing the performance of the on-line services. Deferred Update Replication (DUR) is a well-known approach to design scalable replicated systems. In this method, the database is fully replicated on each distributed node. User threads perform transactions locally and optimistically before a total order is reached. DUR based systems find their best usage when remote transactions rarely conflict. Even in such scenarios, transactions may abort due to local contention on nodes. A generally adopted method to alleviate the local contention is to invoke a local certification phase to check if a transaction conflicts with other local transactions already completed. If so, the given transaction is aborted locally without burdening the ordering layer. However, this approach still results in many local aborts which significantly degrades the performance. The first main contribution of this thesis is PXDUR, a DUR based transactional system, which enhances the performance of DUR based systems by alleviating local contention and increasing the transaction commit rate. PXDUR alleviates local contention by allowing speculative forwarding of shared objects from locally committed transactions awaiting total order to running transactions. PXDUR allows transactions running in parallel to use speculative forwarding, thereby enabling the system to utilize the highly parallel multi-core platforms. PXDUR also enhances the performance by optimizing the transaction commit process. It allows the committing transactions to skip read-set validation when it is safe to do so. PXDUR achieves performance gains of an order of magnitude over closest competitors under favorable conditions. Transactions also form an important part of centralized DBMS, which tend to support multi-threaded access to utilize the highly parallel hardware platforms. The applications can be wrapped in transactions which can then access the DBMS as per the rules of concurrency control. This allows users to develop applications that can run on DBMSs without worrying about synchronization. texttt{Serializability} is the de-facto standard form of isolation required by transactions for many applications. The existing methods employed by DBMSs to enforce serializability employ explicit fine-grained locking. The eager-locking based approach is pessimistic and can be too conservative for many applications. The locking approach can severely limit the performance of DBMSs especially for scenarios with moderate to high contention. This leads to the second major contribution of this thesis is TSAsR, an adaptive transaction processing framework, which can be applied to DBMSs to improve performance. TSAsR allows the DBMS's internal synchronization to be more relaxed and enforces serializability through the processng of external meta-data in an optimistic manner. It does not require any changes in the application code and achieves orders of magnitude performance improvements for high and moderate contention cases. The replicated transaction processing systems require a distributed algorithm to keep the system consistent by ensuring that each node executes the same sequence of deterministic commands. These algorithms generally employ texttt{State Machine Replication (SMR)}. Enhancing the performance of such algorithms is a potential way to increase the performance of distributed systems. However, developing new SMR algorithms is limited in production settings because of the huge verification cost involved in proving their correctness. There are frameworks that allow easy specification of SMR algorithms and subsequent verification. However, algorithms implemented in such framework, give poor performance. This leads to the third major contribution of this thesis Verified JPaxos, a JPaxos based runtime system which can be integrated to an easy to verify I/O automaton based on Multipaxos protocol. Multipaxos is specified in Higher Order Logic (HOL) for ease of verification which is used to generates executable code representing the Multipaxos state changes (I/O Automaton). The runtime drives the HOL generated code and interacts with the service and network to create a fully functional replicated Multipaxos system. The runtime inherits its design from JPaxos along with some optimizations. It achieves significant improvement over a state-of-art SMR verification framework while still being comparable to the performance of non-verified systems.
  • Thumbnail Image
    Taming the Contention in Consensus-Based Distributed Systems
    Arun, Balaji; Peluso, Sebastiano; Palmieri, Roberto; Losa, Giuliano; Ravindran, Binoy (IEEE, 2021-11-01)
    Contention plays a crucial role in the design of consensus protocols. State-of-the-art solutions optimize their performance for either very low or high contention situations. We propose Caesar, a novel multi-leader Generalized Consensus protocol, most suitable for geographical replication, that is optimized for low-to-moderate contention. With an evaluation study, we show that Caesar outperforms other multi-leader (e.g., EPaxos) and single-leader (e.g., Multi-Paxos) competitors by up to 1.7x and 3.5x, respectively, in the presence of 30 percent conflicting requests, in a geo-replicated setting. Furthermore, we acknowledge that there is no one-size-fits- all consensus solution, especially for all levels of contentious workloads. Thus, we also propose Spectrum, a consensus framework that is able to switch consensus protocols at runtime to enable a dynamic reaction to changes in the workload and deployment characteristics. We show empirically that Spectrum can guarantee high availability even during periods of transition between consensus protocols.