News Stay informed about the latest enterprise technology news and product updates.

Disaggregated server benefits lie in efficiency gains

Disaggregated servers are unlikely to go 'boom' in the market the way hyper-convergence has but figure to grow at a more leisurely pace.

The role of a disaggregated server in data centers of the future remains uncertain, as prospective users see its...

potential to increase efficiencies but still with plenty of concerns.

The technology holds the possibility for service providers and enterprise IT shops to make the most of its infrastructure. It also could be just another way for vendors to sell more data center hardware.

The concept of disaggregation involves compute, memory and storage each in a tray, all in discrete units that are compiled, aggregated and composed in a pool instead of individual server units.

"You are treating them all separately and depending on what the app needs, you pull in the right mix in the right proportion," said Jed Scaramella, research director for servers and data center at IDC. He led a roundtable discussion about disaggregated server and composable infrastructure ideas at the recent IDC Directions event in Boston.

For example, provisioning can be done based on whether a workload is compute- or storage-intensive and then scaled incrementally. Service and hosting providers with dynamic workloads will fuel investment in the technology, Scaramella said.

"The concepts have been around for quite a long time, but I don't think the workload demand has been there to fuel investment," he said. There's "absolutely" a use for disaggregated systems by cloud service providers, said Donna Nitchie, an independent contractor retired from IBM who sought more information about the technology at the IDC Directions event. There will also be a use for enterprises that want to keep data on premises and disaggregated servers could be used as the backbone of a private cloud infrastructure, she said.

The payoff is around when you decompose things and drop things back into the pool to be reprovisioned for another application.
Jed Scaramellaresearch director for servers and data center, IDC

"I'm curious about the whole pendulum swing," she said. "It seems like it was going toward integrated and now it is going potentially toward disaggregated. Is this a last ditch effort from the hardware vendors to create value?"

The importance of hardware is declining, Nitchie said, raising the question of just how appealing disaggregated systems will be to the broader user base.

"It might be in high-performance computing and the other areas where they continue to focus on the hardware," she said. The biggest difference users will see first is between systems from original design manufacturers (ODMs) versus original equipment manufacturers (OEMs),  Scaramella said.

Development of the technology is being led by ODMs such as Inspur Group Co. Ltd., Quanta Computer Inc. and Huawei Technologies Co. Ltd., although many ODMs don't have the same capabilities with infrastructure software as do OEMs like EMC, Dell and Cisco, he pointed out.

Scaramella characterized today's disaggregated server options as still "gen zero," including the Intel Rack Scale Architecture platform upon which partners such as Huawei, Quanta and Dell are building.

For OEMs, Cisco UCS M-Series is along the same lines and other modular platforms include Dell FX2 and Hewlett Packard Enterprise Synergy, which is composable but not disaggregated.

"Once you explain it people get it that they can provision things more discretely," he said, noting that when memory runs out users don't have to add another server and can add just add more memory on the fly.

If a workload incrementally scales, disaggregated systems are overkill, Scaramella said. More appropriate workloads would be mobility and online gaming.

"The payoff is around when you decompose things and drop things back into the pool to be reprovisioned for another application," he said.

Expedient Data Centers in Pittsburgh currently uses tools -- some experimental and others as part of partnerships -- that create "intelligence" that helps the company understand the profiles of workloads over time and place them accordingly, said Jonathan Rosenson, vice president of quality assurance and strategic initiatives.

The company tailors hardware to meet the demands of varying workloads as well as to match demand with the supply on the company's tens of thousands of VMs -- and that might mean buying disaggregated systems to serve cloud computing and managed service customers for Expedient's 11 data centers.

"The products (and services) we sell today are dependent on choices we have made and architecture we have in place," Rosenson said. "But that is always evolving and we are in our seventh iteration so anything is possible."

The company has come up with what he claims is a cost-effective infrastructure from multiple vendors that is replicated across its data centers and is scalable.

"If we locked into a single vendor, that may limit us," he said, about disaggregated systems.

Analysts are getting a glimpse into what buyers see as the possible uses for disaggregated systems. However, few if any attended Scaramella's recent session, with the audience mostly vendor representatives and even investors checking out the technology. Similarly, the IDC Directions event in San Jose two weeks earlier included "a lot of vendors and a few users" looking to hear about disaggregated systems.

"The vendors themselves are asking what it is." Scaramella said. "If it matures and keeps going you can start treating everything separately," he said, but today the RAM and CPU have not been separated on the server because there is too much latency.

To pull apart CPU and memory will take silicon photonics, which involves data transfer using optical rays.

Scaramella interviewed 15 potential customers in December, asking them how they think disaggregated systems could benefit their company and some of the challenges they see with it. None of them were using it and the survey focused mainly on service providers, financial services and healthcare companies.

Four of the surveyed users said they could see benefits from disaggregated systems; five said they couldn't see any benefits; six said they needed to understand the technology better.

"More than a third of them were in the wait-and-see mode, so that is pretty indicative of where customers' mindsets are," Scaramella said.

Healthcare providers were the one area that universally said they didn't see a use for disaggregated systems, according to the IDC survey.

Service providers that look to use every asset down to the CPU and treat the data center as a factory floor could also use disaggregated systems to some advantage, he said.

"Whatever they can squeeze out of every piece of hardware, that goes right to their profit," he said.

Enterprises are currently interested in converged infrastructure and hyper-converged infrastructure because both help reduce operating expenses by making staff more efficient and by easing deployment along with simple and streamlined management. Putting disaggregated systems next to CI and HCI raises a lot of questions about interoperability and the two types of hardware working together.

"I see them coexisting even though some people see it as an evolution," Scaramella said. Even though the workloads are different, in many cases, forecasters almost always overestimate the evolution curve and most new hardware finds a sustainable niche, he said. The move to disaggregated systems in the data center will be evolutionary and not revolutionary.

"It will not be like hyper-convergence where it came in and was like 'boom' and disrupted things," Scaramella said. "It is an evolution of architecture, not a revolution."

Robert Gates covers data centers, data center strategies, server technologies, converged and hyper-converged infrastructure and open source operating systems for SearchDataCenter. Follow him on Twitter @RBGatesTT or email him at rgates@techtarget.com.

Next Steps

Get a true sense of Si photonics in the data center

Ten tips on server hardware advances

The new buyers looking into hyper-convergence

Dig Deeper on Server hardware strategy

PRO+

Content

Find more PRO+ content and other member only offers, here.

Join the conversation

3 comments

Send me notifications when other members comment.

By submitting you agree to receive email from TechTarget and its partners. If you reside outside of the United States, you consent to having your personal data transferred to and processed in the United States. Privacy

Please create a username to comment.

What are the biggest advantages, and concerns, you see with disaggregated servers in the data center?
Cancel
The issue with disaggregating memory from CPU is that over the last decade the CPUs have been very tightly coupled not just physically (within 20cm / 6 inches) but also logically to the exact timing of a bank of DRAMs which that CPU (and only that CPU) controls. Sharing memory requires not just distance (with which photonics helps), but also breaking that logical coupling. None of this is rocket science -- multi-socket (multi processor) servers have had to do this for 50 years -- it just requires processor ASICs with different design objectives than the ones which can be put in ODM servers today.

This is also intertwined with persistent, storage class memory, which is emerging into the mainstream this year. While storage class memory could be connected over PCIe in the short term, real server memory needs the lower latency (and other attributes) of connection to the CPU's memory pipeline, not its I/O pipeline.

@FStevenChalmers
(speaking for self, works for Hewlett Packard Enterprise)
Cancel
If I understand disaggregated correctly, I become concerned about the data transfer rates of time-sensitive data (read that: memory). What processor is going to manage the I/O? Will there be a controller in front of the memory components? Top notch vendors have broadened the bandwidth and shortened the response time to memory using IOPs, cache coherency and in some ways NUMA. I can't see how disaggregated servers improves I/O.
Cancel

-ADS BY GOOGLE

SearchWindowsServer

SearchServerVirtualization

SearchCloudComputing

Close