Optically Connected Memory for Disaggregated Data Centers

Gonzalez, Jorge; Gazman, Alexander; Hattink, Maarten; Palma, Mauricio G.; Bahadori, Mehdi N.; Rubio-Noriega, Ruth E.; Orosa, Lois; Glick, Madeleine; Mutlu, Onur; Bergman, Keren; Azevedo, Rodolfo

doi:10.1109/sbac-pad49847.2020.00017

Cited by 8 publications

(5 citation statements)

References 49 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Following the trend for hardware specialization and the desire to better utilize resources as systems scale up, resource disaggregation across the system or a group of racks has been actively researched and deployed in commercial hyperscale datacenters in Google, Facebook, and others [17,55,64]. In addition, many studies focus on disaggregation of GPUs [35] and memory capacity [33,67].…”

Section: Resource Disaggregationmentioning

confidence: 99%

See 1 more Smart Citation

A Case For Intra-rack Resource Disaggregation in HPC

Klenk

Cook

Teh

et al. 2022

ACM Trans. Archit. Code Optim.

Self Cite

View full text Add to dashboard Cite

The expected halt of traditional technology scaling is motivating increased heterogeneity in high performance computing (HPC) systems with the emergence of numerous specialized accelerators. As heterogeneity increases, so does the risk of underutilizing expensive hardware resources if we preserve today’s rigid node configuration and reservation strategies. This has sparked interest in resource disaggregation to enable finer-grain allocation of hardware resources to applications. However, there is currently no data-driven study of what range of disaggregation is appropriate in HPC. To that end, we perform a detailed analysis of key metrics sampled in NERSC’s Cori, a production HPC system that executes a diverse open-science HPC workload. In addition, we profile a variety of deep learning applications to represent an emerging workload. We show that for a rack (cabinet) configuration and applications similar to Cori, a central processing unit (CPU) with intra-rack disaggregation has a 99.5% probability to find all resources it requires inside its rack. In addition, ideal intra-rack resource disaggregation in Cori could reduce memory and NIC resources by 5.36% to 69.01% and still satisfy the worst-case average rack utilization.

show abstract

Section: Resource Disaggregationmentioning

confidence: 99%

“…This capability is referred to as resource disaggregation. In datacenters, resource disaggregation has increased the utilization of GPUs and memory [19,33,35,50,64,67]. Such approaches usually employ a full-system solution where resources can be pooled from across the system.…”

Section: Introductionmentioning

confidence: 99%

A Case For Intra-rack Resource Disaggregation in HPC

Klenk

Cook

Teh

et al. 2022

ACM Trans. Archit. Code Optim.

Self Cite

View full text Add to dashboard Cite

show abstract

“…The O-NIC leverages work on photonic interconnects at Columbia University. 3,7 Figure 1 provides an overview of the FLEET hardware architecture. Briefly, 1 shows how multiple clusters can be interconnected via optical switches, with 3 Tbps aggregate system-to-system throughput in the FLEET Generation 1 system (Gen1) and 12 Tbps in FLEET Generation 2 (Gen2).…”

Section: Hardware Overviewmentioning

confidence: 99%

FLEET—Fast Lanes for Expedited Execution at 10 Terabits: Program Overview

Douglis¹,

Robertson²,

Berg³

et al. 2021

IEEE Internet Comput.

Self Cite

View full text Add to dashboard Cite

“…Most of the works on optically interconnected CDCs focus on the composition of single disaggregated servers to host any service requested by the tenant (e.g., [1,3]), without considering the requirements of deploying disaggregated multi-server (MS) infrastructures. To achieve MS provisioning, in addition to the construction of multiple disaggregated servers, inter-server communications must be also provided.…”

Section: Introductionmentioning

confidence: 99%

On the Complexity of Configuration and Orchestration for Enabling Disaggregated Server Provisioning in Optical Composable Data Centres

Pages¹,

Agraz²,

Spadaro³

2022

J. Opt. Commun. Netw.

View full text Add to dashboard Cite

Due to the limitations of traditional data center (DC) architectures, the concept of infrastructure disaggregation has been proposed. DC resources are separated into multiple blades to be exploited independently. As a result, composable DC (CDC) infrastructures are achieved, enhancing the modularity of resource provisioning. However, disaggregation introduces additional challenges that need to be carefully analyzed. One relates to the potential complexity increase on the orchestration and infrastructure configuration that need to be performed when provisioning resources to support services. This aspect is highly influenced by the distribution of resources at the physical infrastructure. As such, when analyzing the performance of a CDC, it becomes essential to also study the related operational complexity of the resource orchestration and configuration phases. Furthermore, the requirements of several tenant services may impose heterogeneous deployments over the shared physical infrastructure in the form of either disaggregated single-server or multi-server distributions. The associated orchestration/configuration cost is again highly influenced by the data plane architecture of the CDC. With these aspects in mind, in this paper, we provide a methodology for analysis of the complexity of resource orchestration for a service deployment and the associated configuration cost in optical CDCs, considering various service deployment setups. A selected set of CDC architectures found in the literature is employed to quantitatively illustrate how the data plane design and service deployment strategies affect the complexity of infrastructure configuration and resource orchestration.

show abstract

Optically Connected Memory for Disaggregated Data Centers

Cited by 8 publications

References 49 publications

A Case For Intra-rack Resource Disaggregation in HPC

A Case For Intra-rack Resource Disaggregation in HPC

FLEET—Fast Lanes for Expedited Execution at 10 Terabits: Program Overview

On the Complexity of Configuration and Orchestration for Enabling Disaggregated Server Provisioning in Optical Composable Data Centres

Contact Info

Product

Resources

About