System-level implications of disaggregated memory

Lim, Kevin; Turner, Yoshio; Santos, João; AuYoung, Alvin; Chang, Jichuan; Ranganathan, Parthasarathy; Wenisch, Thomas F.

doi:10.1109/hpca.2012.6168955

Cited by 156 publications

(102 citation statements)

References 11 publications

Supporting

Mentioning

102

Contrasting

Order By: Relevance

“…have very different cost, performance, and power scaling trends and this constrains their integration. For instance, recent work [16,17] warns of an impending "memory capacity wall" due to the growing imbalance in the peak compute-to-memory-capacity ratio and argues that traditional compute-memory co-location on a single server will not be sustainable. Similarly, upgrading to a new technology (e.g., NVRAM, memristors, or silicon photonics) or incorporating specialized hardware (e.g., GPUs or accelerators for encryption, coding, or regular-expression matching) can be burdensome since it requires reworking the integration process, server form factor planning, and motherboard designs.…”

Section: Introductionmentioning

confidence: 99%

Network support for resource disaggregation in next-generation datacenters

Han

Egi

Panda

et al. 2013

Proceedings of the Twelfth ACM Workshop on Hot Topics in Networks

134

View full text Add to dashboard Cite

Datacenters have traditionally been architected as a collection of servers wherein each server aggregates a fixed amount of computing, memory, storage, and communication resources. In this paper, we advocate an alternative construction in which the resources within a server are disaggregated and the datacenter is instead architected as a collection of standalone resources.Disaggregation brings greater modularity to datacenter infrastructure, allowing operators to optimize their deployments for improved efficiency and performance. However, the key enabling or blocking factor for disaggregation will be the network since communication that was previously contained within a single server now traverses the datacenter fabric. This paper thus explores the question of whether we can build networks that enable disaggregation at datacenter scales.

show abstract

Section: Introductionmentioning

confidence: 99%

Network support for resource disaggregation in next-generation datacenters

Han

Egi

Panda

et al. 2013

Proceedings of the Twelfth ACM Workshop on Hot Topics in Networks

134

View full text Add to dashboard Cite

show abstract

“…The results and insights from earlier works for disaggregated systems as reported in [5,15,19] are largely obtained from simulation, and did not address some of the recent NoSQL workloads such as Giraph and Cassandra. In this paper, we reported prototyping effort for demonstrating rack scale composability using PCIe switch, and experimental results from running NoSQL and big data workloads such as Giraph, MemcacheD, and Cassandra.…”

Section: Related Workmentioning

confidence: 94%

“…Many of these scenarios indicated that CPU capacity is exhausted before memory capacity is reached, therefore leaving a fraction of the memory unused. It is hypothesized that when the memory is placed in shared pools, higher efficiencies may be achieved by composing servers dynamically through carving out the necessary amounts of memory from these shared pools [9,[15][16][17][18][19]. However these isolated analyses overlooked the performance and cost issues of memory disaggregation.…”

Section: Rack Scale Composable Memorymentioning

confidence: 99%

“…Partially composable memory architecture was proposed by Lim et al [9,15] in which each composable compute node retains a smaller size of memory while the rest of the memory is disaggregated and shared remotely. When a compute node requires more memory to perform a task, the hypervisor integrates (or "compose") the local memory and the remote shared memory to form a flat memory address space for the task.…”

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

Composable architecture for rack scale big data computing

Franke

Parris

et al. 2017

Future Generation Computer Systems

View full text Add to dashboard Cite

Keywords:Big data platforms, Composable system architecture, Disaggregated datacenter architecture, composable datacenter, software defined environments, software defined networking. Abstract:The rapid growth of cloud computing, both in terms of the spectrum and volume of cloud workloads, necessitate re-visiting the traditional rack-mountable servers based datacenter design. Next generation datacenters need to offer enhanced support for: (i) fast changing system configuration requirements due to workload constraints, (ii) timely adoption of emerging hardware technologies, and (iii) maximal sharing of systems and subsystems in order to lower costs. Disaggregated datacenters, constructed as a collection of individual resources such as CPU, memory, disks etc., and composed into workload execution units on demand, are an interesting new trend that can address the above challenges. In this paper, we demonstrated the feasibility of composable systems through building a rack scale composable system prototype using PCIe switch. Through empirical approaches, we develop assessment of the opportunities and challenges for leveraging the composable architecture for rack scale cloud datacenters with a focus on big data and NoSQL workloads. In particular, we compare and contrast the programming models that can be used to access the composable resources, and developed the implications for the network and resource provisioning and management for rack scale architecture.

show abstract

“…In all cases we show results normalized in order to (1) better illustrate the comparative performance among them and (2) present a global view of comparative performance across several figures covering all datasets tested [36,34,41]. To this end, in each figure, we divide all the experimental times by the observed maximum in the particular experiment.…”

Section: Methodsmentioning

confidence: 99%

Distributed text search using suffix arrays

Arroyuelo¹,

Bonacic²,

Gil-Costa³

et al. 2014

Parallel Computing

View full text Add to dashboard Cite

Text search is a classical problem in Computer Science, with many data-intensive applications. For this problem, suffix arrays are among the most widely known and used data structures, enabling fast searches for phrases, terms, substrings and regular expressions in large texts. Potential application domains for these operations include large-scale search services, such as Web search engines, where it is necessary to efficiently process intensive-traffic streams of on-line queries. This paper proposes strategies to enable such services by means of suffix arrays. We introduce techniques for deploying suffix arrays on clusters of distributed-memory processors and then study the processing of multiple queries on the distributed data structure. Even though the cost of individual search operations in sequential (non-distributed) suffix arrays is low in practice, the problem of processing multiple queries on distributed-memory systems, so that hardware resources are used efficiently, is relevant to services aimed at achieving high query throughput at low operational costs. Our theoretical and experimental performance studies show that our proposals are suitable solutions for building efficient and scalable on-line search services based on suffix arrays. IntroductionIn the last decade, the design of efficient data structures and algorithms for textual databases and related applications has received a great deal of attention, due to the rapid growth of the amount of text data available from different sources. Typical applications support text searches over big text collections in a client-server fashion, where the user queries are answered by a dedicated server [15]. The server efficiency-in terms of running time-is of paramount importance in cases where the services demanded by clients generate a heavy work load. A feasible way to overcome the limitations of sequential computers is to resort to the use of several computers, or processors, which work together to serve the ever increasing client demands [19].One such approach to efficient parallelization is to distribute the data onto the processors, in such a way that it becomes feasible to exploit locality via parallel processing of user requests, each on a subset of the data. As opposed to shared-memory models, this distributed-memory model provides the benefit of better * Corresponding author. Address: Av. España 1680, Valparaíso, Chile. Phone: +56 2 432 6722. Fax: +56 2 432 6702. , in distributed memory systems, and describes strategies to reduce the inter-processor communication and to improve the load balance at search time. Indexed Text SearchingThe advent of powerful processors and cheap storage has enabled alternative models for information retrieval, other than the traditional one of a collection of documents indexed by a fixed set of keywords. One is the full text model, in which the user expresses its information need via words, phrases or patterns to be matched for, and the information system retrieves those documents containing the user-specified pa...

show abstract

System-level implications of disaggregated memory

Cited by 156 publications

References 11 publications

Network support for resource disaggregation in next-generation datacenters

Network support for resource disaggregation in next-generation datacenters

Composable architecture for rack scale big data computing

Distributed text search using suffix arrays

Contact Info

Product

Resources

About