Shrinking the Trusted Computing Base of Cloud Infrastructure

35 concepts32 papers1 proposal

In modern cloud computing, we rely heavily on virtualization to isolate tenants sharing the same physical hardware. However, this isolation introduces a fundamental security paradox: the very layer responsible for enforcing isolation—the hypervisor or virtual machine monitor (VMM)—is itself a massive, complex software stack. As Nickolai Zeldovich et al. observe in Hardware Enforcement of Application Security Policies Using Tagged Memory, traditional systems suffer from an excessively large Trusted Computing Base (TCB), where millions of lines of kernel code must be completely bug-free to guarantee security.

When a hypervisor incorporates extensive codebases for hardware emulation, legacy drivers, and complex cloud management features, its attack surface expands dramatically (Wuqiong Pan et al. in Improving Virtualization Security by Splitting Hypervisor into Smaller Components). This complexity creates a dangerous single point of failure. As Shukun Liu notes in A Survey: Main Virtualization Methods and Key Virtualization Technologies of CPU and Memory, a security breach at the monitor level grants an attacker access to all guest virtual machines hosted on the physical hardware.

This vulnerability manifests in critical failure modes like virtual machine escape, where an attacker breaks out of an isolated guest environment to seize root access on the host system (Abdulaziz Nasser A AlFraih in Design of a worm isolation and unknown worm monitoring system based on Honeypot; Wei Fan et al. in Elevated Penetration Attack Models of Virtual Machine Escape Based on FSM). Furthermore, a bloated TCB makes formal mathematical verification of security properties completely intractable (Fengzhe Zhang et al. in CloudVisor: retrofitting protection of virtual machines in multi-tenant cloud with nested virtualization; Zhenjiang Qian et al. in HybridHP:A Verified Lightweight Approach to Provide Lifetime Kernel Integrity Surveillance).

To resolve this, a prominent research direction focuses on shrinking, splitting, or bypassing the hypervisor entirely to minimize the TCB.

Architectural Splitting and User-Space Delegation

One intuitive way to shrink the TCB is to refactor the virtualization layer, separating core security isolation mechanisms from complex management features. If a bug in a non-essential component (like a legacy device driver) can compromise the entire host, moving that component out of the privileged kernel mitigates the risk.

In Improving Virtualization Security by Splitting Hypervisor into Smaller Components, Wuqiong Pan et al. demonstrate this via SplitVisor, an architecture that isolates core security mechanisms from the bloated management layer. Taking this a step further, Jiahao Chen et al. introduce "delegated virtualization" in Security and Performance in the Delegated User-level Virtualization, shifting the bulk of hypervisor functions into user-space. By deprivileging these components, a vulnerability in a hypervisor function results in a standard user-space crash rather than a catastrophic host kernel compromise.

Alternatively, researchers have leveraged nested virtualization to introduce a tiny, trusted security layer beneath the untrusted hypervisor. In CloudVisor: retrofitting protection of virtual machines in multi-tenant cloud with nested virtualization, Fengzhe Zhang et al. propose inserting a minimalist security monitor directly onto the hardware, underneath the commodity hypervisor. The large, complex hypervisor still manages resources and handles VM orchestration, but the underlying monitor intercepts critical operations to protect tenant data privacy and integrity, effectively removing the bloated hypervisor from the security-critical TCB.

Hardware-Assisted Isolation and Confidential Computing

Instead of refactoring the hypervisor, another paradigm aims to eliminate the need to trust the host software stack entirely by using hardware-enforced isolation.

At the application level, hardware enclaves like Intel SGX allow sensitive code to execute securely even on a fully compromised host. For example, Florian Kelbert et al. present SecureCloud: Secure Big Data Processing in Untrusted Clouds, which shields big data processing micro-services within enclaves, removing the cloud provider's hypervisor and operating system from the attack surface. Similarly, Saba Eskandarian et al. develop Fidelius: Protecting User Secrets from Compromised Browsers, combining hardware enclaves with a trusted I/O path to protect user secrets from complex, vulnerable software stacks like modern web browsers.

Modern hardware extensions scale this concept up from individual applications to entire virtual machines. As demystified by Pau-Chen Cheng et al. in Intel TDX Demystified: A Top-Down Approach, Intel Trust Domain Extensions (TDX) use hardware-managed memory encryption and a specialized software module to completely remove the hypervisor from the TCB of guest workloads.

On edge devices, Sandra Deepthy Siby et al. leverage ARM's Confidential Computing Architecture (CCA) in GuaranTEE: Towards Attestable and Private ML with CCA to run machine learning models within secure, isolated "realms." However, the authors note a critical limitation of this hardware-centric approach: while it guarantees data confidentiality and integrity against a malicious hypervisor, a privileged host can still execute denial-of-service attacks by withholding physical resources.

Minimalist Monitors and Bare-Metal Alternatives

A final strategy involves bypassing traditional virtualization layers altogether or replacing them with minimalist software layers designed from the ground up for formal verification. As Jie Qiang argues in Research on the Trusted Enhancement Technology Based on Bare Metal Model Virtual Machine, relying on a full host operating system for virtualization inherently introduces an excessively large TCB.

To eliminate this layer entirely, Amin Mosayyebzadeh et al. design Bolted in A Secure Cloud with Minimal Provider Trust. Bolted is a bare-metal cloud architecture that completely removes the shared hypervisor layer. Instead, it combines hardware-based attestation with a minimal, open-source firmware called Heads, allowing tenants to establish secure private enclaves directly on commodity bare-metal hardware without trusting a provider's virtualization mechanisms.

When a software monitor is required, keeping the codebase small enough for formal verification is paramount. In Sanctorum: A lightweight security monitor for secure enclaves, Ilia A. Lebedev et al. present Sanctorum, a minimalist security monitor for RISC-V processors. By implementing enclave isolation in just roughly 1,000 lines of core C code—as opposed to the millions of lines found in standard hypervisors—the authors demonstrate that a drastically reduced TCB can make formal verification of security properties practically achievable.

Key Foundational Work

CloudVisor: retrofitting protection of virtual machines in multi-tenant cloud with nested virtualization (2011) — contributed 1 concept elements to this direction.
Hardware Enforcement of Application Security Policies Using Tagged Memory (2008) — contributed 1 concept elements to this direction.
History & citationVersion 1 · 2026-10-04

Current · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_trusted_computing_base_expansion_6d2ece61,
  title = {trusted computing base expansion},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/trusted-computing-base-expansion/rdr_6d2ece617b6f4539b10494b474fcdbe6},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-19},
}

This exact version (version 1, frozen text):

@misc{lacuna_trusted_computing_base_expansion_6d2ece61_v1,
  title = {trusted computing base expansion},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/trusted-computing-base-expansion/rdr_6d2ece617b6f4539b10494b474fcdbe6/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-19},
}
About

These directions overlap with this one through shared papers.

See something wrong? Report a mistake