News: 1645538474

  ARM Give a man a fire and he's warm for a day, but set fire to him and he's warm for the rest of his life (Terry Pratchett, Jingo)

Marvell, Dell team on SmartNIC for 5G servers

(2022/02/22)


Dell and Marvell have linked arms to create an accelerator card for servers used in 5G networks.

Announced today ahead of the Mobile World Congress (MWC) conference, the "Dell Open RAN Accelerator Card" uses Marvell's [1]Octeon – silicon dedicated to networking chores – so that the CPUs housed in PowerEdge servers used in 5G networks are freed from processing Layer 1 traffic, and can instead focus on running the Open RAN.

Freeing up the CPU for critical workloads and leaving an accelerator to handle plumbing is the same reasoning the likes of Intel, Nvidia, VMware, Pensando, and most hyperscale clouds offer for their decision to add accelerators to servers under the name SmartNIC, Infrastructure Processing Unit or Data Processing Unit. Qualcomm and HPE [2]launched a very similar accelerator last week .

[3]

In one regard, this is the way 5G was supposed to happen – the standard encourages disaggregation rather than reliance on single vendors and closed standards.

[4]

[5]

On the other hand, developing accelerators is recognition that the vision of x86-powered servers satisfying every niche is not entirely realistic – as with both Dell and HPE deciding their boxen need help, it is clear that vanilla servers aren't always up to the job.

Dell is at least trying to make those servers easier to wield, as accompanying its Mobile World Congress announcement of the Accelerator is news of the "Dell Bare Metal Orchestrator management software", code that allows wide and rapid deployment of desired server images (provided they're running 5G stacks from VMware, Red Hat or Wind River) to hundreds or thousands of machines. Which is just what's needed once servers are found bolted on to 5G radio sites across a network.

[6]

The accompanying "Bare Metal Orchestrator Modules" allow management of servers across their lifecycle.

Dell has teased this stuff since its [7]October 2021 re-definition of hyperconverged infrastructure as a single-node affair, rather than relying on local clusters.

[8]5G masts will be strapped to lampposts and traffic lights – once £4m project figures out who owns them

[9]Ever-growing volumes of data mean computational storage is becoming crucial for HPC, say boffins at Dell's tech chinwag

[10]Nutanix reshuffles product portfolio into bigger bundles

[11]China signs up 400 million new 5G subscribers in a year, more than doubling user population

Reference implementations have of course been cooked up to make this all work as you'd expect because it's a competent vendor that has noticed the opportunity 5G represents. Sam Saba, Dell's head of telecom for Asia Pacific & Japan, told The Register the company has also hired the expertise it needs to satisfy carrier customers.

Saba said when he joined Dell a year ago, it had around 100 staff in the telco team. That number's now closer to 1,000 and Saba said it should end the year with double that number on the books. The company already has a reference customer for a Dell-powered disaggregated 5G – Singaporean carrier Singtel.

As it happens, Singtel said on Tuesday it will offer its 5G-network-as-a-service, with packaged solutions for different industries, and the ability to consume Singtel-provided edge compute resource.

[12]

That last capability smells a bit like Dell's Validated Design for Services Edge, which was upgraded ahead of MWC. The house that Mike built has also announced "ProDeploy for NFVI" to offer templates that enable faster network function virtualization rollout and management. ®

Get our [13]Tech Resources



[1] https://www.theregister.com/2020/09/25/smartnic_dpu/

[2] https://www.theregister.com/2022/02/18/hpe_qualcomm_5g/

[3] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/networks&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2YhUWupQNUBQ14IrzuTe1agAAAFY&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0

[4] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/networks&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44YhUWupQNUBQ14IrzuTe1agAAAFY&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0

[5] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/networks&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33YhUWupQNUBQ14IrzuTe1agAAAFY&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0

[6] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/networks&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44YhUWupQNUBQ14IrzuTe1agAAAFY&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0

[7] https://www.theregister.com/2021/10/13/dell_edge/

[8] https://www.theregister.com/2022/02/09/5g_mobile_masts_lampposts/

[9] https://www.theregister.com/2022/01/31/dell_computational_storage/

[10] https://www.theregister.com/2022/02/15/nutanix_product_changes/

[11] https://www.theregister.com/2022/01/21/chiina_5g_progress/

[12] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/networks&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33YhUWupQNUBQ14IrzuTe1agAAAFY&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0

[13] https://whitepapers.theregister.com/



Is it me?

Anonymous Coward

Or does there seem an awful lot of potential attack vectors in all of these layers?

Firstly, ILO, (on HPE servers, HPE's management system) which I've never found particular robust or stable or easy to manage. Maybe the result of the Active Health Management System that excessively wrote to the 4M flash storage and caused premature hardware failure of HP Server's on-board flash storage.

HPE have not helped, they have consistently just batted away the issue (their support is a minefield to navigate, as is updating ILO, formatting it via scripting (not that easy), even just following the advisories etc. Hours upon hours to nothing. The biggest bugbear with so many failed servers due to ILO, was it was initially impossible to update the system ROMPaq on devices out of 1yr warranty, you had to pay to get access to that update.

Only once the issue was identified by HPE, were fixes made available, several years into the problem, long after the flash memory had been damaged beyond repair for most users. The fixes, that have been put out there, seem superficial 'made to look like they work, but don't', HP added "Format Flash" option, (a botch if there ever was) within ILO interface that just doesn't work (yet to find a machine that it did with the affected fault). Fundamentally, it doesn't solve the issue of the hardware onboard flash storage being excessively written by the Health Management System Engine and damaged beyond repair.

Then there is the underlying iffy Intel Linux based management engine, that in effect works across core network and outer RAN designs. Newer Intel chips having a new Linux based sub-system, that seems, could work across core networks, by getting under the radar.

If Government or others, want to write a complete spying sorry 'security' engine to sit on top of a forked/customised OpenRAN implementation, there really isn't much stopping them with this hardware, it's not that the Qualcomm hardware has holes 'as such' (too early to tell), it just has all the acceleration features to enable holes via customised software. The edge cards seem even sold, going by the blurb, with such marketing, incorporated into those Qualcomm Edge Cards, for on-the-fly decryption.

Maybe it is me, but there seem an awful lot of potential attack vectors here in these products, and HP's ILO, really doesn't help things. Dell seem to re-use rebadged Qualcomm cards, from the information on the Qualcomm site.

I suppose the only way we'll find out, is the hard way.

Re: Is it me?

Anonymous Coward

To add: (because this is easier than dealing with HPE support).

HPE, you really should put out a recall, and offer mainboard replacements for free to all affected products with ILO issues. (Gen8/Gen9) "Controller firmware revision 2.10.00 NAND write failture: Media is in a WRITE-PROTECTED state".

The issue clearly wasn't acted upon, and dealt with within a time-frame that the end-user could have done anything to prevent the systematic hardware failure of the onboard flash, due to the excessive amount of data being written by the Active Health System Management Engine.

I won't be recommending your products until you do.

A girl spent a couple hours on the phone talking to her two best
friends, Maureen Jones, and Maureen Brown. When asked by her father why she
had been on the phone so long, she responded "I heard a funny story today
and I've been telling it to the Maureens."