Online GNN inference has been widely explored by applications such as online recommendation and financial fraud detection systems, where even minor delays can result in significant financial impact. Real-time dynamic graph sampling enables online GNN inference to reflect the latest graph updates in real-world graphs. However, online GNN inference typically demands millisecond-level latency Service Level Objectives (SLOs) as its performance guarantees, which poses great challenges for existing dynamic graph sampling approaches based on graph databases. The issues mainly arise from two aspects: long tail latency due to imbalanced data-dependent sampling and large communication overhead incurred by distributed sampling. To address these issues, we propose Helios, an efficient distributed dynamic graph sampling service to meet the stringent latency SLOs. The key ideas of Helios are 1) pre-sampling the dynamic graph in an event-driven approach, and 2) maintaining a query-aware sample cache to build the complete K-hop sampling results locally for inference requests. Experiments on multiple datasets show that Helios achieves up to 67× higher serving throughput and up to 32× lower P99 query latency compared to baselines.

The actor model has been widely adopted in building stateful middle-tiers for large-scale interactive applications, where ACID transactions are useful to ensure application correctness. In this paper, we present Snapper, a new transaction library on top of Orleans, a popular actor system. Snapper exploits the characteristics of actor-oriented programming to improve the performance of multi-actor transactions by employing deterministic transaction execution, where pre-declared actor access information is used to generate deterministic execution schedules. The deterministic execution can potentially improve transaction throughput significantly, especially with a high contention level. Besides, Snapper can also execute actor transactions using conventional nondeterministic strategies, including S2PL, to account for scenarios where actor access information cannot be pre-declared. A salient feature of Snapper is the ability to execute concurrent hybrid workloads, where some transactions are executed deterministically while the others are executed nondeterministically. This novel hybrid execution is able to take advantage of the deterministic execution while being able to account for nondeterministic workloads. Our experimental results on two benchmarks show that deterministic execution can achieve up to 2x higher throughput than nondeterministic execution under a skewed workload. Additionally, the hybrid execution strategy can achieve a throughput that is close to deterministic execution when there is only a small percentage of nondeterministic transactions running in the system.

Graph query services (GQS) are widely used today to interactively answer graph traversal queries on large-scale graph data. Existing graph query engines focus largely on optimizing the latency of a single query. This ignores significant challenges posed by GQS, including fine-grained control and scheduling during query execution, as well as performance isolation and load balancing in various levels from across user to intra-query. To tackle these control and scheduling challenges, we propose a novel scoped dataflow for modeling graph traversal queries, which explicitly exposes concurrent execution and control of any subquery to the finest granularity. We implemented Banyan, an engine based on the scoped dataflow model for GQS. Banyan focuses on scaling up the performance on a single machine, and provides the ability to easily scale out. Extensive experiments on multiple benchmarks show that Banyan improves performance by up to three orders of magnitude over state-of-the-art graph query engines, while providing performance isolation and load balancing.

In existing big data stacks, the processes of analytical processing and knowledge serving are usually separated in different systems. In Alibaba, we observed a new trend where these two processes are fused: knowledge serving incurs generation of new data, and these data are fed into the process of analytical processing which further fine tunes the knowledge base used in the serving process. Splitting this fused processing paradigm into separate systems incurs overhead such as extra data duplication, discrepant application development and expensive system maintenance. In this work, we propose Hologres, which is a cloud native service for hybrid serving and analytical processing (HSAP). Hologres decouples the computation and storage layers, allowing flexible scaling in each layer. Tables are partitioned into self-managed shards. Each shard processes its read and write requests concurrently independent of each other. Hologres leverages hybrid row/column storage to optimize operations such as point lookup, column scan and data ingestion used in HSAP. We propose Execution Context as a resource abstraction between system threads and user tasks. Execution contexts can be cooperatively scheduled with little context switching overhead. Queries are parallelized and mapped to execution contexts for concurrent execution. The scheduling framework enforces resource isolation among different queries and supports customizable schedule policy. We conducted experiments comparing Hologres with existing systems specifically designed for analytical processing and serving workloads. The results show that Hologres consistently outperforms other systems in both system throughput and end-to-end query latency.

Fault-tolerance techniques for stream processing engines can be categorized into passive and active approaches. However, both approaches have their own inadequacies in Massively Parallel Stream Processing Engines (MPSPE). The passive approach incurs a long recovery latency especially when a number of correlated nodes fail simultaneously, while the active approach requires extra replication resources. In this paper, we propose a new fault-tolerance framework, which is Passive and Partially Active (PPA). In a PPA scheme, the passive approach is applied to all tasks while only a selected set of tasks will be actively replicated. The number of actively replicated tasks depends on the available resources. If tasks without active replicas fail, tentative outputs will be generated before the completion of the recovery process. We also propose effective and efficient algorithms to optimize a partially active replication plan to maximize the quality of tentative outputs. We implemented PPA on top of Storm, an open-source MPSPE and conducted extensive experiments using both real and synthetic datasets to verify the effectiveness of our approach.

Correlated failures that usually involve a number of nodes failing simultaneously have significant effect on systems' availability, especially for streaming applications that require real-time analysis. Most state-of-the-art distributed stream processing engines focus on recovering individual operator failure. By analyzing the existing recovery techniques, we identify the challenges and propose a fault-tolerance framework that can tolerate both individual and correlated failures with minimum overhead during the system's normal execution. Our progressive and query-centric recovery paradigm carefully schedules the recovery of failed operators based on the current availability of resources, such that the outputs of queries can be recovered as early as possible. We also formulate the new problem of recovery scheduling under correlated failures and design algorithms to optimize the recovery latency with a performance guarantee.

Correlated failures in large-scale clusters have significant effects on systems’ availability, especially for streaming data applications that run continuously and require low processing latency. Most stateof- the-art distributed stream processing engines (DSPEs) adopt a blocking recovery paradigm, which, upon correlated failure, would block the progress of recovery until sufficient new resources for recovery are available. As the arrival of new resources is usually progressive, a blocking paradigm fails to minimize the recovery latency. To address this problem, we propose a progressive and query-centric recovery paradigm where the recovery of the failed operators would be carefully scheduled to progressively recover the outputs of queries as early as possible based on the current availability of resources. In this work, we propose and implement a fault-tolerance framework which supports progressive recovery after correlated failures with minimum overhead during the system’s normal execution. We also formulate the new problem of recovery scheduling under correlated failures and design effective algorithms to optimize the recovery latency. The proposed methods are implemented on Apache Storm and preliminary experiments are conducted to verify their validity

Fault-tolerance techniques for stream processing engines can be categorized into passive and active approaches. A typical passive approach periodically checkpoints a processing task's runtime states and can recover a failed task by restoring its runtime state using its latest checkpoint. On the other hand, an active approach usually employs backup nodes to run replicated tasks. Upon failure, the active replica can take over the processing of the failed task with minimal latency. However, both approaches have their own inadequacies in Massively Parallel Stream Processing Engines (MPSPE). The passive approach incurs a long recovery latency especially when a number of correlated nodes fail simultaneously, while the active approach requires extra replication resources. In this paper, we propose a new fault-tolerance framework, which is Passive and Partially Active (PPA). In a PPA scheme, the passive approach is applied to all tasks while only a selected set of tasks will be actively replicated. The number of actively replicated tasks depends on the available resources. If tasks without active replicas fail, tentative outputs will be generated before the completion of the recovery process. We also propose effective and efficient algorithms to optimize a partially active replication plan to maximize the quality of tentative outputs. We implemented PPA on top of Storm, an open-source MPSPE and conducted extensive experiments using both real and synthetic datasets to verify the effectiveness of our approach.

In this paper, we consider the problem of continuous dissemination of time series data, such as sensor measurements, to a large number of subscribers. These subscribers fall into multiple subscription levels, where each subscription level is specified by the bandwidth constraint of a subscriber, which is an abstract indicator for both the physical limits and the amount of data that the subscriber would like to handle. To handle this problem, we propose a system framework for multi-scale time series data dissemination that employs a typical tree-based dissemination network and existing time-series compression models. Due to the bandwidth limits regarding to potentially sheer speed of data, it is inevitable to compress and re-compress data along the dissemination paths according to the subscription level of each node. Compression would caused the accuracy loss of data, thus we devise several algorithms to optimize the average accuracies of the data received by all subscribers within the dissemination network. Finally, we have conducted extensive experiments to study the performance of the algorithms.