VMware has quietly walked back its SmartNIC ambitions

VMware has quietly stopped selling firewalls for SmartNICs. SmartNICs, aka data processing units (DPUs), are network cards that include a modest processor. Hyperscale clouds routinely use the devices to handle security and networking chores, by running network functions as VMs on the SmartNIC’s processor. Cloud operators like …

[Bebu sa Ware]

Virtzilla … didn’t excite customers

At this point only a notice in the corporate obituary column might evoke any such response.

[Anonymous Coward]

Re: Virtzilla … didn’t excite customers

Every extra check you put in the I/O path to “make things better” has a cost in overhead.

When you're dealing with millisecond response times, piling on crap like this needs a better justification than another security feature on a sales sheet. It may earn its keep when it catches an attack, but you pay the performance bill every day.

You're already running security agents, AV, FW, ZTNA and now AI-assisted monitoring. Keep stacking them up and you can end up buying more processing power and memory just to handle the same useful workload. The overhead becomes part of the hardware budget, and somehow that's called progress.

Show us the latency under load, the throughput hit, and the false-positive rate. Then we can judge whether the additional protection is worth the cost.

Layered defenses also need to account for how quickly you can respond when something gets through. Security tends to put more hurdles in front of the “good guys.” That becomes particularly galling when an attacker exploits a zero-day vulnerability and bypasses the controls you're still obliged to follow. If they've gained privileged access, they could be tearing through your systems while you're still waiting for an MFA code so you can get in and stop them.

And if a hacker gets onto your storage VLAN, gains control of the storage, and can destroy your recovery copies as well, you are boned regardless. Proper isolation, restricted administrative access, and backups you've actually tested still matter, but the time frames to respond are shrinking daily.

Slapping “AI” on another layer doesn't make its overhead disappear.

[Anonymous Coward]

Hey, look euver there!

I seem to recall that concessions on SmartNIC/DPU support were what got Broadcom's takeover of VMware over the EU finishing line. This news makes me wonder, as I did at the time, whether this was a diversionary tactic.

[spireite]

Hasn't it got more to do with the utter lack of trust that customers have in Broadcom feeding into this?

[Sproggit]

Bingo.

Came here to make essentially the same comment... What with the Tesco court case - among others - one has to suspect that Broadcom have a bit of a toxic brand on their hands right now. Companies who were thinking of doing business with them have looked at the way they are treating their VMWare clients and have perhaps been thinking that getting hitched to a strategy with a hard dependency on this company is a bad idea.

So that "Customers have been talking but not buying" might very well be true.

But if the customers' "talk" has been to say, "Good-bye", then that's a very different matter altogether.

When a company starts to see a drop-off in revenue and interest, one of the first thing it does is rapidly scale back on fringe research, narrow-margin or low-volume products or anything that takes energy away from their core revenue generators. Which would absolutely fit the visible fact pattern we see here.

[Anonymous Coward]

Hypercrap

Hyperscale. Hyperconverged. Hyper... whatever. I recall when a large computer/storage manufacturer for whom I worked said that "Hyper<insert something here>" is THE WAY of the future, and all other architectures will drop off the face of the earth. Didn't happen, isn't going to happen. Virtualization certainly has its place, but so does independent processing. And clustering. And <insert technology here>. When will this industry finally learn that one size does NOT fit all?

[AtomicWombat]

Re: Hypercrap

~20 years ago, IBM sold us their new "Storage Tank" SAN File System which was billed an an all-singing, all-dancing, spinning-rims-and-ground-effects storage solution that would do everything we could ever possibly want in an early 2000s HPC environment. We were also told that GPFS, their previous product, was going the way of the dodo.

However, as we got closer to deployment, lots of weird little caveats starting coming out. SAN Volume Controller suddenly became a requirement for no discernable reason. iSCSI became a requirement as well, and this was back when iSCSI was a barely functional tech preview.

Finally, we were in a planning meeting where our senior admin declared his intent to put users' home director is on it, and the IBM folks reacted as if he had sprouted a second head and hadn't been saying this for almost a year. They sheepishly admitted that it wasn't likely to work for that, and that all their other customers (which numbered on the single digits) were hosting DB2 on it.

SPOILER ALERT: SANFS was quietly scrubbed from IBM's product list, and it's interesting features were added to GPFS

[mikus]

Fear the deprecation

I'm sure they fear everyone, including nvidia of trying to obsolete vmware, so of course. Hide your head in the sand, they won't get you.