EE Seminar: Scalable Storage Acceleration: A Parallel Queues Approach
Electrical Engineering Systems Seminar
הרישום לסמינר יתבצע באמצעות המודל ,באחריותכם לודא כי יש לכם שם משתמש וסיסמא לכניסה למודל מהנייד
Speaker: Dhruv Wagh
M.Sc. student under the supervision of Prof. Guy Even and Prof. Boaz Patt-Shamir
Sunday, 6th April 2025, at 15:00
Room 011, Kitot Building, Faculty of Engineering
Scalable Storage Acceleration: A Parallel Queues Approach
Abstract
The rapid growth of data intensive applications has led to an increase in disparity between the processing speed and data access speeds often resulting in huge bottlenecks at an IO level. Historically, the primary slowdown was due to the inherent speed of the storage device, however with the introduction of high performance SSDs, the bottleneck now primarily lies in accessing and processing the data.
This thesis addresses the performance gap between data processing and storage access in high-performance computing. We propose a storage accelerator that leverages near-data processing and computational storage to improve efficiency. Our proposed design reduces data transfer overhead on the processor and enables computation closer to the data source. It is a special purpose hardware processing unit that connects a host to an array of disks. From the point of view of the host, it appears as a single storage device with a valid filesystem. Performance scales linearly with the number of drives (k), achieving a speedup of S = k theoretically. This work addresses system design, OS interaction, and performance analysis, with significant implications for high-data computing.
השתתפות בסמינר תיתן קרדיט שמיעה לתלמידי תואר שני ושלישי = עפ"י רישום שם מלא + מספר ת.ז. בדף הנוכחות שיועבר באולם במהלך הסמינר
