Search arXiv⌕ Search
Home/Archive/arXiv

arXiv · 2603.09023

The Missing Memory Hierarchy: Demand Paging for LLM Context Windows

Tony Mason
View original source ↗Read source document ↗Export metadata ↓

Abstract

The context window of a large language model is not memory. It is L1 cache: a small, fast, expensive resource that the field treats as the entire memory system. There is no L2, no virtual memory, no paging. Every tool definition, every system prompt, and every stale tool result occupies context for the lifetime of the session. The result is measurable: across 857 production sessions and 4.45 million effective input tokens, 21.8% is structural waste. We present Pichay, a demand paging system for LLM context windows. Implemented as a transparent proxy between client and inference API, Pichay interposes on the message stream to evict stale content, detect page faults when the model re-requests evicted material, and pin working-set pages identified by fault history. In offline replay across 1.4 million simulated evictions, the fault rate is 0.0254%. In live production deployment over 681turns, the system reduces context consumption by up to 93% (5,038KB to 339KB); under extreme sustained pressure, the system remains operational but exhibits the expected thrashing pathology, with repeated fault-in of evicted content. The key observation is that the problems the field faces, such as context limits, attention degradation, cost scaling, lost state across sessions, are virtual memory problems wearing different clothes. The solutions exist: working set theory (Denning, 1968), demand paging, fault-driven replacement policies, and memory hierarchies with multiple eviction-managed levels. We describe the architecture of a full memory hierarchy for LLM systems (L1 through persistent storage), report on the first three levels deployed in production use (L1 eviction, L2 fault-driven pinning, L3 model-initiated conversation compaction), and identify cross-session memory as the remaining frontier.

Explore related subjects

cs.OScs.AIcs.SE

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Tony Mason. 2026-03-09. The Missing Memory Hierarchy: Demand Paging for LLM Context Windows. https://arxiv.org/abs/2603.09023

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

Don't Let AI Agents YOLO Your Files: Information and Control in Agent-Native Filesystems

AI coding agents regularly misuse their filesystem access, causing data corruption, loss, and leakage. We conduct the first systematic study of this problem through an analysis of 290 public reports. Our study reveals two fundamental gaps: users and agents have limited information about filesystem effects and insufficient control over them. To close these gaps, we propose to shift information and control from agents to filesystems. We introduce agent-native filesystems and identify three primitives they should provide: introspect effects, undo mutations, and gate accesses. These primitives let agents operate autonomously while reserving user interaction for sensitive accesses and final review. We build YoloFS, an agent-native filesystem. YoloFS stages mutations until the user commits them, snapshots intermediate states for agent self-correction, and uses progressive permission to let users adapt access rules during execution. We evaluate YoloFS with a new methodology that captures interactions among the user, agent, and filesystem. On 11 tasks with hidden side effects, YoloFS enables agents to self-correct in 8 and stages all mutations for user review. On 112 routine tasks, YoloFS reduces user interaction while matching the baseline success rate. YoloFS is open-sourced at https://github.com/YoloFS/YoloFS.

cs.OS↗
arXiv2026

Netkit: Specializing Linux Packet Delivery for Container Networks

Cloud-native microservices architectures rely on network namespaces for isolation, with the overhead of container communications remaining a critical performance bottleneck. While colocating containers on the same host mitigates some of this overhead, it cannot match the performance of communication within a single network namespace. Existing solutions either require application rewrites or fail to support the full Linux network stack expected by containerized applications. In this paper, we present netkit, an eBPF-based datapath that specializes the Linux networking stack to eliminate redundant backlog queue traversals during network namespace transitions. netkit leverages eBPF to transparently redirect packets between namespaces, bypassing unnecessary buffering while preserving compatibility with existing container applications. Our implementation in the Linux kernel, integrated with minimal changes to the Cilium network plugin for Kubernetes, improves throughput by up to 37\% and achieves parity between container-to-container and process-to-process communications, effectively closing the performance gap introduced by namespace isolation.

cs.OS↗
arXiv2026

Grouper: Scheduling Groups for Multi-Tenant Microsecond-Scale Microservices

Microsecond-scale core allocation makes colocating latency-critical services with batch work worthwhile. A thread that finds no work parks within microseconds and its core goes to a batch task. Putting one back costs $\sim$18 $μ$s, as the allocator must discover that a core is wanted and then take it from the batch task holding it. A monolith pays that tax once per request, a microservice chain pays it at every hop in both directions, and a multi-tenant host multiplies it again, because every tenant's hops queue at the same allocator. On our port of DeathStarBench's hotelReservation, going from two tenants to ten takes a hop from 39 to 222 $μ$s and a 10-RPC path's median from 456 to 2,445 $μ$s, a fivefold degradation even though no tenant's own load changed. We introduce Grouper and the scheduling group, a set of isolated runtimes that the allocator treats as one allocation and accounting unit, whose members may hand cores directly to one another. A service sending an RPC donates its core to the peer through an unprivileged kernel fast path, so the core follows the request through the call graph. The allocator retains control through reconciliation, core-addressed revocation and a pooled budget but leaves the critical path; its load falls from $Θ(R \cdot H)$ to $Θ(R)$ in request rate $R$ and hop count $H$. Over a grid of two to ten tenants at 1,000-30,000 requests per second each, Grouper outperforms Caladan (the allocator Junction also builds on) and Linux by up to 7.9$\times$ and 3.4$\times$ at the median and 4.1$\times$ and 14.2$\times$ at the tail, and leaves batch work more throughput than Caladan at over 70% of load points.

cs.OS↗