RAID sucks (and what we can do about it)
CEME 1210 | Fri 07 Aug 11:45 a.m.–12:30 p.m.
Presented by
-
Christoph Hellwig has been working on the Linux kernel and various other Free Software projects since 1999.
Since 2004 he has been running his own consulting business specializing in Linux file systems and storage, but also working on virtualization and the RISC-V open instruction set. He is an active participant in the NVMe technical working group and the IEFT Network File System Version 4
working group. He is one of the maintainers of the Linux NVMe driver, and one of the busiest contributors to the Linux kernel.
Christoph Hellwig has been working on the Linux kernel and various other Free Software projects since 1999.
Since 2004 he has been running his own consulting business specializing in Linux file systems and storage, but also working on virtualization and the RISC-V open instruction set. He is an active participant in the NVMe technical working group and the IEFT Network File System Version 4 working group. He is one of the maintainers of the Linux NVMe driver, and one of the busiest contributors to the Linux kernel.
Abstract
Redundant Array of Independent Disks (RAID) has long been used to store large amounts of data while protecting against data loss
and improving performance over that of a single disk.
Naive RAID implementations always have had problems providing full data integrity in the face of system crashes or power failures, and cannot keep up with the increased performance of modern SSDs. Furthermore, no solution exists today to handle the sequential write constraint of zoned storage (e.g. SMR HDDs) which makes up an ever-increasing share of shipped storage capacity.
This talk unveils a new solution that supports full data integrity and supports all modern storage devices based on existing file system code in the Linux kernel.
Redundant Array of Independent Disks (RAID) has long been used to store large amounts of data while protecting against data loss and improving performance over that of a single disk.
Naive RAID implementations always have had problems providing full data integrity in the face of system crashes or power failures, and cannot keep up with the increased performance of modern SSDs. Furthermore, no solution exists today to handle the sequential write constraint of zoned storage (e.g. SMR HDDs) which makes up an ever-increasing share of shipped storage capacity.
This talk unveils a new solution that supports full data integrity and supports all modern storage devices based on existing file system code in the Linux kernel.