Proceedings of the 22nd International Middleware Conference最新文献

Lachesis: a middleware for customizing OS scheduling of stream processing queries Lachesis:用于自定义流处理查询的操作系统调度的中间件

Proceedings of the 22nd International Middleware Conference Pub Date : 2021-12-06 DOI: 10.1145/3464298.3493407

Dimitris Palyvos-Giannas, G. Mencagli, M. Papatriantafilou, Vincenzo Gulisano

{"title":"Lachesis: a middleware for customizing OS scheduling of stream processing queries","authors":"Dimitris Palyvos-Giannas, G. Mencagli, M. Papatriantafilou, Vincenzo Gulisano","doi":"10.1145/3464298.3493407","DOIUrl":"https://doi.org/10.1145/3464298.3493407","url":null,"abstract":"Data streaming applications in Cyber-Physical Systems enable high-throughput, low-latency transformations of raw data into value. The performance of such applications, run by Stream Processing Engines (SPEs), can be boosted through custom CPU scheduling. Previous schedulers in the literature require alterations to SPEs to control the scheduling through user-level threads. While such alterations allow for fine-grained control, they hinder the adoption of such schedulers due to the high implementation cost and potential limitations in application semantics (e.g., blocking I/O). Motivated by the above, we explore the feasibility and benefits of custom scheduling without alterations to SPEs but, instead, by orchestrating the OS scheduler (e.g., using nice and cgroup) to enforce the scheduling goals. We propose Lachesis, a standalone scheduling middleware, decoupled from any specific SPE, that can schedule multiple streaming applications, run in one or many nodes, and possibly multiple SPEs. Our evaluation with real-world and synthetic workloads, several SPEs and hardware setups, shows its benefits over default OS scheduling and other state-of-the-art schedulers: up to 75% higher throughput, and 1130x lower average latency once such SPEs reach their peak processing capacity.","PeriodicalId":154994,"journal":{"name":"Proceedings of the 22nd International Middleware Conference","volume":"12 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-12-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131827064","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 5

CAT: content-aware tracing and analysis for distributed systems CAT:分布式系统的内容感知跟踪和分析

Proceedings of the 22nd International Middleware Conference Pub Date : 2021-12-06 DOI: 10.1145/3464298.3493396

Tânia Esteves, Francisco Neves, Rui Oliveira, J. Paulo

{"title":"CAT: content-aware tracing and analysis for distributed systems","authors":"Tânia Esteves, Francisco Neves, Rui Oliveira, J. Paulo","doi":"10.1145/3464298.3493396","DOIUrl":"https://doi.org/10.1145/3464298.3493396","url":null,"abstract":"Tracing and analyzing the interactions and exchanges between nodes is fundamental to uncover performance, correctness and dependability issues almost unavoidable in any complex distributed system. Existing monitoring tools acknowledge this importance but, so far, restrict tracing to the external attributes of I/O messages, thus missing a wealth of information in them. We present CaT, a non-intrusive content-aware tracing and analysis framework that, through a novel similarity-based approach, is able to comprehensively trace and correlate the flow of network and storage requests from applications. By supporting multiple tracing tools, CaT can balance the coverage of captured events with the impact on applications' performance. The conducted experimental evaluation considering two widely used applications (TensorFlow and Apache Hadoop) shows how CaT can improve the analysis of distributed systems. The results also exemplify the trade-offs that can be used to balance tracing coverage and performance impact. Interestingly, in certain cases, full coverage of events can be attained with negligible performance and storage overhead.","PeriodicalId":154994,"journal":{"name":"Proceedings of the 22nd International Middleware Conference","volume":"77 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-12-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123942977","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 6

SwitchFlow

Proceedings of the 22nd International Middleware Conference Pub Date : 2021-12-02 DOI: 10.1145/3464298.3493391

Xiaofeng Wu, J. Rao, Wei Chen, Hang Huang, Chris H. Q. Ding, Heng Huang

{"title":"SwitchFlow","authors":"Xiaofeng Wu, J. Rao, Wei Chen, Hang Huang, Chris H. Q. Ding, Heng Huang","doi":"10.1145/3464298.3493391","DOIUrl":"https://doi.org/10.1145/3464298.3493391","url":null,"abstract":"Accelerators, such as GPU, are a scarce resource in deep learning (DL). Effectively and efficiently sharing GPU leads to improved hardware utilization as well as user experiences, who may need to wait for hours to access GPU before a long training job is done. Spatial and temporal multitasking on GPU have been studied in the literature, but popular deep learning frameworks, such as Tensor-Flow and PyTorch, lack the support of GPU sharing among multiple DL models, which are typically represented as computation graphs, heavily optimized by underlying DL libraries, and run on a complex pipeline spanning CPU and GPU. Our study shows that GPU kernels, spawned from computation graphs, can barely execute simultaneously on a single GPU and time slicing may lead to low GPU utilization. This paper presents SwitchFlow, a scheduling framework for DL multitasking. It centers on two designs. First, instead of scheduling a computation graph as a whole, SwitchFlow schedules its subgraphs and prevents subgraphs from different models to run simultaneously on a GPU. This results in less interference and the elimination of out-of-memory errors. Moreover, subgraphs running on different devices can overlap with each other, leading to a more efficient execution pipeline. Second, SwitchFlow maintains multiple versions of each subgraph. This allows subgraphs to be migrated across devices at a low cost, thereby enabling low-latency preemption. Results on representative DL models show that SwitchFlow achieves up to an order of magnitude lower tail latency for inference requests collocated with a training job.","PeriodicalId":154994,"journal":{"name":"Proceedings of the 22nd International Middleware Conference","volume":"65 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-12-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126153764","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 6

Highly-available and consistent group collaboration at the edge with colony 高度可用和一致的群体协作在边缘与群体

Proceedings of the 22nd International Middleware Conference Pub Date : 2021-12-02 DOI: 10.1145/3464298.3493405

Ilyas Toumlilt, P. Sutra, M. Shapiro

引用次数: 4

Gossip consensus 八卦的共识

Proceedings of the 22nd International Middleware Conference Pub Date : 2021-12-02 DOI: 10.1145/3464298.3493395

Daniela Cason, Nenad Milosevic, Zarko Milosevic, F. Pedone

引用次数: 4

FaaSTCC: efficient transactional causal consistency for serverless computing FaaSTCC:用于无服务器计算的高效事务因果一致性

Proceedings of the 22nd International Middleware Conference Pub Date : 2021-12-02 DOI: 10.1145/3464298.3493392

Taras Lykhenko, João Rafael Pinto Soares, Luís Rodrigues

{"title":"FaaSTCC: efficient transactional causal consistency for serverless computing","authors":"Taras Lykhenko, João Rafael Pinto Soares, Luís Rodrigues","doi":"10.1145/3464298.3493392","DOIUrl":"https://doi.org/10.1145/3464298.3493392","url":null,"abstract":"In this paper we study mechanisms that permit to augment the FaaS middleware with support for Transactional Causal Consistency (TCC). At first glance, it may seem that offering TCC to FaaS applications can trivially be achieved, given that the FaaS paadigm does not prevent applications from selecting the storage service with the properties they need. Unfortunately, most TCC storage services ensure consistency only to individual client processes, while a FaaS application is executed by multiple, independent, worker processes. Therefore, there is the need to coordinate the workers, a task that can be a significant source of overhead. We propose a novel architecture to support TCC in FaaS, named FaaSTCC, that significantly reduces the coordination overhead. FaaSTCC achieves this goal by augmenting the workers with a caching layer and by implementing novel mechanisms that maximize the cache usage. First, our storage layer offers to the caching layer a promise, that sets a horizon where the versions retrieved by the cache are guaranteed to be consistent. Second, in FaaSTCC, functions coordinate using snapshot intervals, that support the lazy identification of the read snapshot, increasing the chances of using the cached values. We have implemented and experimentally evaluated FaaSTCC. Our results show that FaaSTCC achieves up to 5x lower average latency and 6x lower tail latency than previous work.","PeriodicalId":154994,"journal":{"name":"Proceedings of the 22nd International Middleware Conference","volume":"13 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-12-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115703481","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 11

RamCast

Proceedings of the 22nd International Middleware Conference Pub Date : 2021-12-02 DOI: 10.1145/3464298.3493393

Long Hoang Le, Mojtaba Eslahi-Kelorazi, Paulo R. Coelho, Fernando Pedone

引用次数: 5

Experience Paper: Towards enhancing cost efficiency in serverless machine learning training 经验论文:提高无服务器机器学习训练的成本效率

Proceedings of the 22nd International Middleware Conference Pub Date : 2021-12-02 DOI: 10.1145/3464298.3494884

Marc Sánchez Artigas, Pablo Gimeno Sarroca

引用次数: 14

Lachesis 拉克西斯

Proceedings of the 22nd International Middleware Conference Pub Date : 2021-12-02 DOI: 10.14271/dms-12009-de

Dimitris Palyvos-Giannas, G. Mencagli, M. Papatriantafilou, Vincenzo Gulisano

引用次数: 0

Snarl: entangled merkle trees for improved file availability and storage utilization 缠绕:缠绕的默克尔树，提高文件可用性和存储利用率

Proceedings of the 22nd International Middleware Conference Pub Date : 2021-12-02 DOI: 10.1145/3464298.3493397

Racin Nygaard, Vero Estrada-Galiñanes, H. Meling

{"title":"Snarl: entangled merkle trees for improved file availability and storage utilization","authors":"Racin Nygaard, Vero Estrada-Galiñanes, H. Meling","doi":"10.1145/3464298.3493397","DOIUrl":"https://doi.org/10.1145/3464298.3493397","url":null,"abstract":"In cryptographic decentralized storage systems, files are split into chunks and distributed across a network of peers. These storage systems encode files using Merkle trees, a hierarchical data structure that provides integrity verification and lookup services. A Merkle tree maps the chunks of a file to a single root whose hash value is the file's content-address. A major concern is that even minor network churn can result in chunks becoming irretrievable due to the hierarchical dependencies in the Merkle tree. For example, chunks may be available but can not be found if all peers storing the root fail. Thus, to reduce the impact of churn, a decentralized replication process typically stores each chunk at multiple peers. However, we observe that this process reduces the network's storage utilization and is vulnerable to cascading failures as some chunks are replicated 10X less than others. We propose Snarl, a novel storage component that uses a variation of alpha entanglement codes to add user-controlled redundancy to address these problems. Our contributions are summarized as follows: 1) the design of an entangled Merkle tree, a resilient data structure that reduces the impact of hierarchical dependencies, and 2) the Snarl prototype to improve file availability and storage utilization in a real-world storage network. We evaluate Snarl using various failure scenarios on a large cluster running the Ethereum Swarm network. Our evaluation shows that Snarl increases storage utilization by 5X in Swarm with improved file availability. File recovery is bandwidth-efficient and uses less than 2X chunks on average in scenarios with up to 50% of total chunk loss.","PeriodicalId":154994,"journal":{"name":"Proceedings of the 22nd International Middleware Conference","volume":"108 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-12-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122747826","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 4