Endangered Virtualization Species Forced to Evolve

The conflicting demands of security and flexibility have forced the evolution of basic virtualization tools such that they play more nicely with high-performance computing -- while maintaining virtualization's lofty security goals.

Virtualized software stacks bear inherent performance and latency issues; this has led some pundits to predict that traditional virtual switches (vSwitches) and hypervisors could be on their way to extinction. (See Unknown Document 740150.)

While virtualization-driven isolation may be great for security, as with all problems of security versus user-friendliness, the issue remains one of balance. (See Uber Loses Customer Data: Customers Yawn & Keep Riding.)

"One of the interesting challenges with [virtualization] is that if you wish to get maximum flexibility in a nice big set of cloud resources, you have a nice single hypervisor everything-in-together kind of model; now that is the 'all your eggs in one basket' problem," related Alex Leadbeater, a compliance executive at BT and vice chair of the NFV Security Group at the European Telecommunications Standards Institute (ETSI) , to Security Now. "If you ever got a single zero-day exploit on the hypervisor or if anybody ever managed to break in period, you'd lose everything."

Conversely, Leadbeater pointed out, micro-separation of hypervisors can lead to inflexibility as resources potentially get distributed to blocks from which they can't necessarily be readily accessed as needed.

(Source: Pixabay)

"In order to isolate resources, we actually have to have multiple root administrators all fighting in parallel," continued Leadbeater. "The [issue] is one of the key building blocks that the [network operator] industry has to work its way out of."

Leadbeater acknowledged that some methods alternative to traditional vSwitching have eased some of these flexibility burdens while helping to maintain secure isolation -- but it still remains to be seen what "the best way of doing it" is. While deep-packet acceleration -- such as Intel's Data Path Development Kit (DPDK) -- has helped the intrinsic flexibility issues, sometimes offloading data directly onto the bare metal server itself or other means of direct hardware access may prove desirable for optimum flexibility -- potentially re-introducing security concerns. And these security concerns are heightening as time goes by as tenant isolation becomes ever more critical to tomorrow's network.

Therefore, according to Ildiko Vancsa, ecosystem technical lead at the OpenStack Foundation, the additional layers of security and tenant isolation that vSwitches and hypervisors provide mean they're not likely to go away anytime soon -- and that, as network needs have evolved, these technologies have thus had to themselves evolve.

"Next-generation vSwitching, such as FD.io, is able to play a role of providing tenant isolation and traffic optimization for a variety of workload types -- leveraging software switching as the complexity of multi-tenant workloads in the network increases will be important," Vancsa told Security Now. "Looking at [new vSwitching and hypervisor] technologies, one can deploy workloads with more flexibility depending on the context of the deployment with these types of solutions."

In particular, Vancsa pointed to Kata Containers -- an open source project announced by OpenStack at December's OpenStack Summit in Sydney -- which purports to make virtualized infrastructures faster and more efficient by offering "the ability to run container management tools directly on bare metal without sacrificing workload isolation."

Traditionally, containers have had to be run within hypervisors to offer the same kind of heightened isolation that unikernels -- stripped-down, single-purpose mini-VMs unto themselves -- do. (See Unknown Document 740150.)

Consequently, if OpenStack's claims are true, the Kata Containers project gives container technology a huge competitive boost in security against the more esoteric yet far more isolated solution of unikernels (although unikernels generally possess a smaller attack surface by their very -- minimalistic -- nature). The further upshot is that such developments could well pave the way for traditional hypervisors to be phased out of networks someday.

Emphasis on "someday."

Other relatively nascent virtualization tools are meanwhile helping to make both hypervisors and containers more secure and more efficient. Leadbeater, for instance, referenced the potential for Intel's Software Guard Extensions (SGX) -- which can purportedly keep security-critical applications protected in an encrypted "enclave" even if the underlying platform or virtual instance itself has been compromised.

In any case, Leadbeater indicated that ETSI will be addressing best methodologies for these virtualized-network security concerns over the course of 2018.

"Are there tweaks around the edges that we can make that give us fundamental security improvements by tweaking the platforms on which [a virtualized instance] is sitting?" asked Leadbeater rhetorically. "Some of that may require hardware changes. Some of it may require software changes. Some of it may require changes in the way in which the policy management's done."

Related posts:

—Joe Stanganelli, principal of Beacon Hill Law, is a Boston-based attorney, corporate-communications and data-privacy consultant, writer, and speaker. Follow him on Twitter at @JoeStanganelli.