[Colloquium] Reminder: Li/MS Presentation/May 9, 2018

[Margaret Jaffey via Colloquium]

This is a reminder about Huaicheng's MS Presentation tomorrow.

------------------------------------------------------------------------------
Date:  Wednesday, May 9, 2018

Time:  2:30 PM

Place:  Ryerson 277

M.S. Candidate:  Huaicheng Li

M.S. Paper Title: The CASE of FEMU: Cheap, Accurate, Scalable and
Extensible Flash Emulator

Abstract:
Cheap and extensible research platforms are a key ingredient in
fostering wide-spread Solid-State Drives (SSDs) research. While SSD
simulators are popular, they only support in-SSD research. SSD
hardware development platforms are expensive and experience wear-out
issues. This leaves software-based flash emulators an excellent
alternative choice. Unfortunately, the state of existing emulators is
bleak; they are either outdated, non-scalable, or not open-sourced.

We argue that it is a critical time for storage research community to
have a new software-based emulator. To this end, we present FEMU, a
QEMU-based flash emulator, with the following four ``CASE'' benefits.

FEMU is a software (QEMU-based) flash emulator for fostering future
full-stack software/hardware SSD research. FEMU is cheap
(open-sourced), relatively accurate (0.5-38\% variance as a drop-in
replacement of OpenChannel SSD), scalable (can support 32 parallel
channels/chips), and extensible (support internal-only and split-level
SSD research).

First, FEMU is \textbf{\underline{c}}heap (\$0) as it is an
open-sourced software. FEMU has been successfully used in several
projects, some of which appeared in top-tier OS and storage
conferences. We hope FEMU will be useful to broader communities and
accelerate research in broader areas.

Second, FEMU is (relatively) \textbf{\underline{a}}ccurate. For
example, FEMU can be used as a drop-in replacement for OpenChannel
SSD; thus, future research that extends LightNVM can be performed on
top of FEMU with relatively accurate results (\eg, \errMin-\errMax\%
variance in our tests). With FEMU, prototyping SSD-related kernel
changes can be done without a real device.

Third, FEMU is \textbf{\underline{s}}calable. As we optimized the QEMU
stack with various techniques, such as exitless interrupt and skipping
QEMU AIO components, FEMU can scale to 32 IO threads and still achieve
a low latency (as low as \latMin\us\ under a \latGHZ GHz CPU). As a
result, FEMU can accurately emulate 32 parallel channels/chips,
without unintended queueing delays.

Finally, FEMU is \textbf{\underline{e}}xtensible. Being a QEMU-based
emulator, FEMU can support internal-SSD research (only FEMU layer
modification), kernel-only research such as software-defined flash
(only Guest OS modification on top of unmodified FEMU), and
split-level research (both Guest OS and FEMU modifications). FEMU also
provides many new features not existent in other emulators, such as
OpenChannel and multi-device/RAID support, extensible interfaces via
NVMe commands, and page-level latency variability.

Huaicheng's advisor is Prof. Haryadi Gunawi

Login to the Computer Science Department website for details:
 https://www.cs.uchicago.edu/phd/ms_announcements#huaicheng

=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
Margaret P. Jaffey            margaret at cs.uchicago.edu
Department of Computer Science
Student Support Rep (Ry 156)               (773) 702-6011
The University of Chicago      http://www.cs.uchicago.edu
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=