Please note: This PhD seminar will be given online.
Emil
Tsalapatis, PhD
candidate
David
R.
Cheriton
School
of
Computer
Science
Supervisor: Professor Ali Mashtizadeh
Ensuring consistency between an application’s ephemeral state in memory and persistent state on disk is a source of significant bugs even in mature applications. In the first part of the talk we present the Aurora single level store, an OS that simplifies persistence by automatically persisting all traditionally ephemeral application state. Aurora continuously persists complex applications ranging from databases to web browsers 100 times per second with modest overhead. By persisting whole applications and providing ways to manipulate their state, Aurora facilitates novel mechanisms that would otherwise be prohibitively difficult to implement. Our work on the Aurora SLS will be presented at the 18th Workshop on Hot Topics in Operating Systems (HotOS XVIII), June 1–3.
In the second part of the talk we present Metropolis, a serverless framework on top of Aurora that uses OS level sharing to combine high storage density and performance. One of the central problems in serverless computing is cold starts, i.e., invocations of functions that have high latency caused by creating a VM or container for the instance. Snapshot based systems create new instances using existing images, reducing cold start times at the cost of significant disk usage. Metropolis uses Aurora’s copy on write storage hierarchy to share resources between otherwise unrelated snapshots to simultaneously achieve low cold start latency, high density in memory, and minimal disk usage. Our system creates functions 7 times faster and uses 40 times less disk space than vHive, the state of the art in snapshot based serverless frameworks.
Bio: Emil Tsalapatis is a 3rd year PhD student working with Ali Mashtizadeh. His research focuses on novel abstractions for modern OSes, cloud and serverless computing, memory and storage hierarchies, and file systems.
To join this PhD seminar on Zoom, please go to https://zoom.us/j/92268050403?pwd=bVZyS2Nmc2QwRGZOQzNSbzBCM3ROUT09.