SmartNICs haven't soared so VMware will allow retrofits in old servers
vSphere to also add support for PCIe cards packed with CXL RAM to give boxes a boost, faster vCenter upgrades, and more
Explore Buyers have not rushed to adopt SmartNICs in their VMware environments, but the virtualization giant will soon support their use in more servers – a move it hopes will spur a new wave of adoption.
SmartNICs – aka data processing units (DPUs) – are network cards that include some compute and storage capacity. Hyperscalers use them to offload networking and security chores that would otherwise occupy cores that could be rented to customers. VMware liked that approach so much that late in 2022 vSphere 8 gained a Distributed Services Engine to allow use of SmartNICs in conventional servers and datacenters.
Paul Turner, product management veep for VMware's Cloud Infrastructure Platform, told The Register on Wednesday that "DPUs are doing OK but it isn't flying out the door."
The main reason sales aren't stronger, he opined, is that VMware and its hardware partners have allowed DPUs to run only in new servers. That's meant adoption is tied to hardware refresh cycles, and tricky economic conditions have meant many organizations are happy to sweat their machines for an extra year or two.
VMware now plans to allow vSphere to support DPUs in old servers – a facility Turner said will arrive in "the next quarter."
Another reason DPUs haven't taken off is that VMware only supports one per server, while customers are fond of twin-NIC rigs. VMware is working to allow twin-DPU support on an as-yet-undetermined timeframe. Support for accelerated storage I/O is also in the works, adding to the current ability to speed other network traffic and run a firewall.
Another addition coming to a future vSphere release is support for a "memory expander" – a PCIe card crammed with memory that meets the Compute Express Link standard and can therefore allow memory pooling. Micron already makes just such a card, packing 256GB of DRAM.
As Turner explained to The Register, plenty of orgs would like servers with more memory, but have used all available DIMM slots so can't just add more chips. Buying new servers is expensive, because that means buying new memory too – and memory is often the most expensive component.
Replacing a server's memory with higher-capacity modules is also costly – and wasteful, unless orgs have another destination ready for the RAM.
- VMware’s vSphere 8 Update 1 debuts under revised product release regime
- VMware edges towards its multi-cloud vision, with a side quest to virtualize AI
- VMware turns 25 today: Is it a mature professional or headed back to Mom's house?
- VMware adds subscription version of basic vSphere for server consolidation
Memory expanders can save the day by offering a way to add RAM to an existing server, with CXL allowing it to appear to the CPU as part of a pool of addressable memory.
VMware has a head start on this sort of thing thanks to Project Capitola – a pooled memory tech it worked on with Intel. It was supposed to create interest in the Optane persistent memory tech that Chipzilla abandoned in July 2022.
Turner said the work VMware did for Capitola has been reused preparing for work on CXL support, which is already present in the ESX kernel and will soon likely emerge as support for memory expanders.
Before that lands, vSphere users can look forward to a second update: to version 8 of the server virtualization suite. Turner said the release will shrink the update time required for vCenter, the server management component of vSphere, from 30 to three minutes.
He also teased a boost to vSphere's automation abilities thanks to support for Terraform and adding "suitable REST interfaces to everything."
"We really expect people to be able to run this as infrastructure as code, and be able to run at scale," Turner said.
Improved certificate management is another addition, among other security enhancements. Turner said these were added partly as ongoing improvements and partly because VMware recognizes that its hypervisors and associated infrastructure are mission-critical for customers – making any serious security flaw in its products a crisis. ®