Graduation Semester and Year
Spring 2026
Language
English
Document Type
Dissertation
Degree Name
Doctor of Philosophy in Computer Science
Department
Computer Science and Engineering
First Advisor
Dr. Song Jiang
Second Advisor
Dr. Hui Lu
Third Advisor
Dr. Ashraf Aboulnaga
Fourth Advisor
Dr. Bradley Settlemyer
Abstract
Modern computing systems increasingly run on diverse hardware platforms and support applications with widely different access patterns, performance goals, and data lifecycles. In this setting, traditional one-size-fits-all approaches to memory and storage management are often inefficient because they apply fixed policies regardless of application behavior, workload context, or hardware asymmetry. Such generic designs can lead to unnecessary data movement, wasted bandwidth, excessive rewriting, poor resource utilization, and degraded user-perceived performance. This dissertation is motivated by the view that optimal memory and storage management should be application-driven: instead of treating all data uniformly, systems should adapt their decisions to how applications actually access, reuse, and prioritize data. Across both memory and storage, the central goal is to reduce avoidable overheads while preserving or improving performance. \\
In the area of memory management, this dissertation studies how application behavior can guide more efficient use of limited or heterogeneous memory resources. In smartphone systems, MemSaver shows that background application memory should not be managed solely by recency-based policies or aggressive killing under pressure; instead, event-specific access history can be used to predict which pages are likely to be needed during future app switches, allowing memory to be reclaimed while preserving near-ideal responsiveness. In heterogeneous server memory systems, MemProphet extends the same application-driven principle to DRAM–CXL environments, where data placement decisions must account for tier-dependent latency and bandwidth. By combining fast emulation with workload-aware access analysis, it enables rapid identification of suitable allocation policies and reveals which application regions are more tolerant of placement in slower memory tiers. Together, these works demonstrate that memory systems become substantially more effective when allocation, retention, and migration decisions are guided by application access patterns rather than static, uniform heuristics. \\
In the area of storage management, this dissertation focuses on reducing unnecessary data rewriting in write-optimized key-value stores. Conventional LSM-tree designs rely on fixed level-by-level compaction policies that treat all data movement similarly, even when workload structure makes such rewriting wasteful. CollapseDB challenges this rigid approach by showing that key-range density varies across levels and can be exploited to choose more efficient compaction targets and destinations. By enabling opportunistic multi-level compaction, it reduces write amplification, lowers I/O overhead, and improves throughput without sacrificing the core benefits of LSM-based storage. This work illustrates the broader thesis argument on the storage side: data movement should not be dictated solely by static structural rules, but should instead be informed by workload characteristics and the actual cost of moving data. In this way, the dissertation advances a unified perspective in which both memory and storage systems achieve better efficiency and performance by replacing one-size-fits-all management with application-driven adaptation. \\
Keywords
Memory Management, DRAM, CXL, LSM-Tree, Compaction, application-aware memory management, CXL, tiered memory, data movement optimization, workload characterization, mobile memory, storage systems, LSM-tree compaction, write amplification, resource efficiency
Disciplines
Computer and Systems Architecture | Data Storage Systems
License
This work is licensed under a Creative Commons Attribution 4.0 International License.
Recommended Citation
Challa, Venkata Naga Prajwal, "TOWARDS APPLICATION-DRIVEN OPTIMAL MEMORY AND STORAGE MANAGEMENT" (2026). Computer Science and Engineering Dissertations. 9.
https://mavmatrix.uta.edu/cse_dissertations2/9