The Write-Anywhere-File- Layout (WAFL) - NetApp Community

36 Pages·2012·1.72 MB·English

The Write-Anywhere-File-
Layout
(WAFL)
Ohad Rodeh
TheWrite-Anywhere-File-Layout(WAFL)(cid:150)p.1/36
Introduction
This lecture is based on
File System Design for an NFS File
Dave Hitz, James Lau, Michael
Server Appliance
Malcolm. Proceedings of the USENIX Winter 1994
Technical Conference
The WAFL design is used today in Network-Appliance
(cid:2)lers
TheWrite-Anywhere-File-Layout(WAFL)(cid:150)p.2/36
Appliance for NFS
NFS is Network File System
1. A standard protocol to access a remote (cid:2)le-system
2. Create/Delete (cid:2)le
3. Read/Write
An appliance is a special purpose device
TheWrite-Anywhere-File-Layout(WAFL)(cid:150)p.3/36
Main ideas
New (cid:2)le-system idea
1. Write-Anywhere-File-Layout (WAFL)
2. Create snapshots ef(cid:2)ciently
3. Uses a copy-on-write technique
4. Minimizes disk-space that snapshots consume
Snapshots are used to eliminate the need for
(cid:2)le-system consistency checks after an unclean
shutdown
TheWrite-Anywhere-File-Layout(WAFL)(cid:150)p.4/36
Design goals
WAFL is to be used inside an NFS appliance
Tuned speci(cid:2)cally for NFS
Increased write workload due to problematic caching
at client
WAFL should:
1. Provide fast NFS service
2. Support large (cid:2)le systems (tens of GB) that grow
dynamically as disks are added. This simpli(cid:2)es
management, users don’t want multiple partitions.
3. Provide high performance while supporting RAID
4. Support RAID in order to tolerate failed disks
TheWrite-Anywhere-File-Layout(WAFL)(cid:150)p.5/36
gracefully
5. Restart quickly after a crash. Traditional FSCK is
too slow.
Snapshots
Snapshot: read-only copy of the entire (cid:2)le system.
Why use snapshots?
1. Users can access snapshots through NFS to
recover (cid:2)les that they have accidentally changed or
removed
2. System administrators can use snapshots to create
backups safely from a running system
3. WAFL uses Snapshots internally so that it can
recover quickly from crashes
TheWrite-Anywhere-File-Layout(WAFL)(cid:150)p.6/36
Structure
WAFL is similar in many ways to FFS
It uses 4 KB blocks with no fragments
The i-node is similar to FFS, with some exceptions
1. Contains 16 block pointers
2. All the block pointers refer to blocks at the same
level
3. I-nodes for (cid:2)les smaller than 64 KB use the 16
block pointers to point to data blocks
4. I-nodes for (cid:2)les larger than 64 KB point to indirect
blocks which point to actual (cid:2)le data
5. I-nodes for even larger (cid:2)les point to doubly indirect
TheWrite-Anywhere-File-Layout(WAFL)(cid:150)p.7/36
blocks
6. For very small (cid:2)les, data is stored in the i-node itself
Meta data (cid:2)les
WAFL stores meta-data in (cid:2)les
There are three meta-data (cid:2)les
1. I-node (cid:2)le, contains the i-nodes for the (cid:2)le system
2. Block-map (cid:2)le, identi(cid:2)es free blocks
3. I-node-map (cid:2)le, identi(cid:2)es free i-nodes
The term map is used instead of bit map because
these (cid:2)les use more than one bit for each entry
TheWrite-Anywhere-File-Layout(WAFL)(cid:150)p.8/36
Why use (cid:2)les to hold
meta-data?
Allows writing meta-data blocks anywhere on disk
Write-anywhere allows operating ef(cid:2)ciently with RAID
1. WAFL works with RAID-4
2. Scheduling multiple writes to the same RAID stripe
whenever possible
3. Avoid the 4-to-1 short-write penalty
Makes it easy to increase the size of the (cid:2)le system on
the (cid:3)y
TheWrite-Anywhere-File-Layout(WAFL)(cid:150)p.9/36
Why use (cid:2)les to hold
meta-data? II
When a new disk is added
1. Server automatically increase the sizes of the
meta-data (cid:2)les
2. The system administrator can increase the number
of i-nodes in the (cid:2)le system manually if the default
is too small
TheWrite-Anywhere-File-Layout(WAFL)(cid:150)p.10/36