Friday, March 29, 2024
Search
  
Monday, January 15, 2018
 IBM Researchers Bring Memory Disaggregation to Data Centers
You are sending an email that contains the article
and a private message for your recipient(s).
Your Name:
Your e-mail: * Required!
Recipient (e-mail): *
Subject: *
Introductory Message:
HTML/Text
(Photo: Yes/No)
(At the moment, only Text is allowed...)
 
Message Text:

Researchers at IBM Research - Ireland are developing a concept and prototype for low-power and high-utilization disaggregated cloud data centres that enable the dynamic creation of fit-for-purpose computing environments from a pool of disaggregated resources.

In today's cloud data centres, the server main-board together with its hardware components form the baseline, monolithic building block that the rest of the data centre hardware/software stack design builds upon. In such high-end conventional systems, the proportionality of IT resources is fixed during design time and remains static throughout each technology refresh cycle. This comes with known ramifications in terms of low system resource utilization, costly upgrade cycles and degraded energy proportionality. IBM Researchers have been trying to make this arrangement more efficient, flexible and agile with these resources.

Put simply, a data centre is a large collection of servers, each offering certain amounts of resources such as CPU cores and memory. Applications deployed to the cloud are decomposed into their basic processes, each with its own CPU and memory requirements. These processes must be started on data centre servers that have enough capacity.

CPU and memory are physically confined together within server boundaries. Because of this, an application process cannot draw resources from more than one server. The effect is fragmentation of spare CPU and memory resources. This means loss in economies of scale and a higher energy footprint for the cloud provider and higher service charges for the user.

The new concept is to break the physical boundaries of data centre servers by disaggregating CPUs and memory into separate physical entities. This transforms the aggregate data centre memory to a single resource pool, from which any CPU can draw resource. As requests arrive, the system can connect cores to memory on-the-fly, building platforms that match exactly their requirements.

These changes to the infrastructure are transparent to applications, so developers can keep building applications with the tools they are familiar with and deploy them to the Cloud as usual. The ultimate advantages of this disaggregation of CPU and memory are manifold. More requests can be served with the same amount of resources and memory-centric applications are no longer limited to the memory of a single server. You can also independently upgrade processors and memory, bringing agility and full modularity to the service provider.

IBM researchers have built a small scale prototype of the system that proves the feasibility of this approach. "So far, our experimental results have been extremely encouraging, forecasting that our architecture will help reducing up to four times the energy that today is spent on unused computing power," according to the researchers.

"We continue working to improve our technology and build larger-scale proof-of-concepts to pave the way to large scale adoption of disaggregated cloud data centers. This work spearheads the innovation in moving from today's server-as-the-unit model to a pooled-computing model; it will enable an arbitrary sizing of disaggregated IT resources that can be deployed where and when required to perfectly match cloud user requirements. This will bring the cloud to unprecedented efficiency levels while promising a drastic reduction on data centres energy footprint," they said.

Results from this work and demonstrations of the prototype will be presented at AISTECS 2018 workshop, part of the 13th HiPEAC Conference on High Performance Embedded Architectures and Compilers held in Manchester, UK, and the DATE 2018 conference and exhibition held in Dresden, Germany.

The project is led by IBM Research Ireland, together with 10 industrial and academic partners across the European Union. The efforts are part of an EU funded project, dReDBox.

 
Home | News | All News | Reviews | Articles | Guides | Download | Expert Area | Forum | Site Info
Site best viewed at 1024x768+ - CDRINFO.COM 1998-2024 - All rights reserved -
Privacy policy - Contact Us .