This project would investigate the storage access patterns of LLM agents using open-source state-of-the-art inference frameworks such as vLLM and SGLang and workload traces such as the ones released by Mooncake, Alibaba, Swiss AI Initiative, etc. Depending on the size and duration of the project, it can also propose or implement optimizations for the storage access. This project would be suitable …
Supervisors: Pınar Tözün, Jens Birk Andersen
Semester: Fall 2026
Tags: SSDs, LLMs, agents, modern storage

Traditionally solid-state drives (SSDs) does not give the users the ability to control the data placement on the SSD. This often leads to suboptimal performance and lowers SSD lifetime, since SSDs internally don’t allow in-place updates. The updated disk pages are written elsewhere and the old versions have to be garbage collected. This poses problems if data with different lifetimes and …
Supervisors: Pınar Tözün
Semester: Fall 2026
Tags: SSDs, data management systems, modern storage

This project would investigate key-value(KV)-cache offloading in large language model (LLM) inference engines, with particular emphasis on data placement and storage efficiency. First, the project would use an emulator such as FEMU to characterize workload access patterns, including hot-key distributions, write intensity, and write amplification factor (WAF). Then, we would evaluate the potential …
Supervisors: Pınar Tözün, Jens Birk Andersen
Semester: Fall 2026
Tags: SSDs, LLMs, kv-cache, modern storage

A recent MSc thesis project at ITU integrated xNVMe to DuckDB. DuckDB, like many popular database systems, rely on a syncronous IO backend and the POSIX filesystem interfaces for its ease of use. xNVMe, in contrast, allows a unified interface for alternative and more performant IO backends such as io-uring and spdk. Thus, integration of xNVMe into DuckDB offers interesting avenues for data …
Supervisors: Pınar Tözün
Semester: Fall 2025
Tags: SSDs, data management systems, modern storage

This rise of hardware accelerators to meet the demand of AI workloads has also led to a variety of novel methods to leverage GPUs for traditional data analytics workloads. A key concern for any data-intensive system using GPUs is the efficiency of moving the data to the accelerator. In this project, we will investigate ways to improve data movement to GPUs by focusing on the steps of the data path …
Supervisors: Pınar Tözün
Semester: Fall 2025
Tags: SSDs, GPU-centric IO, data analytics, modern storage

Traditionally solid-state drives (SSDs) does not give the users the ability to control the data placement on the SSD. This often leads to suboptimal performance and lowers SSD lifetime, since SSDs internally don’t allow in-place updates. The updated disk pages are written elsewhere and the old versions have to be garbage collected. This poses problems if data with different lifetimes and …
Supervisors: Pınar Tözün
Semester: Fall 2024
Tags: SSDs, data management systems, modern storage

In the past decade, data management community has focused on main-memory systems or main-memory-optimized systems. This focus has put the commodity memory hierarchy (DRAM and processor caches) into center when it comes to workload characterization studies. Today, with the evolution of persistent storage technologies such as NVRAM (persistent memory solution of Intel) and NVMe SSDs, data systems …
Supervisors: Pınar Tözün
Semester: Fall 2022
Tags: workload characterization, tracing, modern storage, data-intensive systems