- All mobile network operators want to reduce energy consumption levels in their radio access networks
- There’s a big focus in the industry on how this can be achieved with special software tools
- But there are key developments at the hardware level too, says HPE’s Geetha Ram
As telcos shift towards virtualised radio access network (vRAN) architectures and consider how to meet all of their infrastructure key performance indicators (KPIs), they need to consider not just what amazing advances can be made from the use of the new generation of software tools but also the advances being made in the hardware that underpins next-generation network architectures.
That’s according to a slightly frustrated Geetha Ram, worldwide head of telco compute at Hewlett Packard Enterprise (HPE), who believes that the servers supporting vRAN and Open RAN tests, trials and (increasingly) rollouts are still regarded as commoditised boxes rather than highly engineered and carrier-grade systems.
“When we talk to the operators about TCO, the total cost of ownership, we keep getting beaten up repeatedly on the three Ps of virtual and Open RAN: Power consumption, performance, and price,” said Ram, who added that it’s hard to achieve all three.
“There are many ways to enhance performance: There’s the latest Intel CPU [central processing unit], AMD CPU, a Nokia inline card can do much more, there’s a x100 accelerator from Qualcomm – there are lots of options. But power and price are the tough ones and the more we talk to operators, the more we realise that power consumption almost equates to price because of the rising operating costs,” noted Ram.
Not that this is reflected in the conversations she has with operators – at least, not yet.
“When operators talk about the hardware, they're always talking about price, price and price, and always saying they are too expensive. They’re not looking at all of the other benefits they can get from open hardware, like optimised power consumption,” she explained. “We're trying to change the mindset of operators by explaining this stuff to them, but it’s hard when all we see are articles about RAN optimisation that focus on everything being software driven. People forget there are things you can do in the hardware, but people just commoditise the hardware,” complained the HPE executive, adding that there are plenty of innovative developments ongoing in the hardware layer.
“Maybe not everyone knows this but we have some algorithms that HPE has patented – two patents that have been filed – on making network processes, even sub-processes, within L1 cards dormant when not in use, and to do it dynamically. That’s something we offer with our hardware – it’s not something we offer as a separate software package, this is something that’s built into our hardware. It looks at the traffic coming in on the fronthaul and decides when to turn certain processes on and off, dynamically controlling the power consumption. With in-line [accelerator] card architecture, we’ve seen anything from 15% to 20% power savings straight away” because there are times when you don’t need to keep any of the CPU cores active, she explained.
Those numbers, added Ram, come from tests carried out by two major European operators in their lab. “They have given us feedback, their benchmark results, and there’s a range because it depends on the volume of traffic,” she said.
Ram expects to see these capabilities being deployed in production networks when operators start to deploy servers based on Intel’s 4th generation Xeon processors (Sapphire Rapids) that will be generally available in the first quarter of 2024.
And she wishes that operators focused more on the impact that hardware innovation can have on power consumption.
“Our core competency is hardware – we do this type of thing in our sleep. The hardware vendors can do a lot before anything even hits the software. And the algorithms we’ve developed, they’re available through our Redfish APIs on our servers, so you could have higher layers of software control them, so there’s flexibility,” she noted.
And it’s not just about power management – there’s innovation too in server design with regards to how many PCI-E slots are available, but that’s not something operators tend to consider, said Ram. The current thinking is to just add another server, but that’s not good for TCO, she explained.
There are ways to be more efficient by considering all the hardware parameters. “A box is not a box is not a box… because a box that has one card that supports narrowband [traffic] is different from the box that supports a mid-band three-cell sector. They are not the same,” and many operators are loathe to compare IT hardware with traditional RAN hardware from the major incumbent vendors because they consider all servers to be enterprise hardware, “but the servers we provide for telcos are carrier grade, they are fully NEBS [network equipment building system] compliant,” said Ram.
But it’s an uphill battle for now, it seems, with so much focus on the software side of vRAN and Open RAN developments and the cost of server hardware, and Ram at least admits that the hardware vendors might be partly to blame for that. There was a time when the server vendors were “playing the price game, undercutting each other, instead of stepping back and looking at the TCO.
“There is the optics power consumption and all the actual capex costs and figuring out whether you need five appliances, or just three or two or even one, and looking at the overall performance and then putting the whole TCO picture together and that’s what we have now done. I think this needs to be thought of from a hardware perspective or a system view, not just from software alone,” concluded Ram.
- Ray Le Maistre, Editorial Director, TelecomTV