Zeta file system

Zettabyte File System

Zettabyte File System
Robert Miłkowski System Group Manager Wirtualna Polska

Zettabyte File System

What is ZFS?
● ● ● ● ●

Completely new 128-bit file system Integrated volume manager Best data integrity Simple management Great performance
UNIX DAYS – Gdańsk 2006 2

Zettabyte File System

Traditional file systems
●

No protection for silent datacorruption
●

Can't guarantee correct data

●

Hard to manage
Partitions, slices, labels ● Not flexible storage space management ● Lots of limits ● Platform dependent
●

●

Too much complexity with FS+VM
UNIX DAYS – Gdańsk 2006 3

Zettabyte File System

Wish list
● ● ● ● ●

Make it SCALABLE Make it MANAGABLE Make it DYNAMIC Make it SAFE Make it EASIER
UNIX DAYS – Gdańsk 2006 4Zettabyte File System

Let's do it from the scratch
●

Pooled storage
No need for slices, partitions, volumes, etc. ● Make as many file systems as you want for free
●

●

End-to-end data integrity
●

Guarantee that application gets correct data Always consistent on disk – meta and user data
UNIX DAYS – Gdańsk 2006 5

●

Transactional
●

Zettabyte File System

TraditionalFS+VM
FS FS FS

Volume

Volume

Volume

UNIX DAYS – Gdańsk 2006

6

Zettabyte File System

ZFS pooled storage
ZFS ZFS ZFS

Storage Pool

UNIX DAYS – Gdańsk 2006

7

Zettabyte File System

ZFS pooled storage - example

# zpool create home c5t40d0 c5t40d1 c5t40d2 c5t40d3 # zfs create home/milek # zfs create home/www # zfs create home/milek/mail # zfs set compression=onhome/milek/mail # # zfs list NAME USED AVAIL REFER MOUNTPOINT home 72K 18.0T 10K /home home/milek 18.5K 18.0T 9.50K /home/milek home/milek/mail 9K 18.0T 9K /home/milek/mail home/www 9K 18.0T 9K /home/www #
UNIX DAYS – Gdańsk 2006 8

Zettabyte File System

ZFS Quoas & Reservations
● ● ● ●

You can reserve space for a given fs You can set quota for a given fs quota/reservation forhierarchies No quotas for uid/gid

UNIX DAYS – Gdańsk 2006

9

Zettabyte File System

Data integrity
●

Both data and meta-data are checksumed
●

No silent data corruption

●

Everything is Copy-On-Write
Never overwrite live data ● Always consistent on disk ● No need for fsck-like utility
●

UNIX DAYS – Gdańsk 2006

10

Zettabyte File System

End-to-end data integrity
●Checksum checked after block is in memory
● ●

Whole IO path is checked Corrects driver bugs, phantom writes, etc.

●

Checksum and data block stored separately
● ●

Checksum is stored in parent block Entire pool is self-validating

●

Protects from accidental overwrites
UNIX DAYS – Gdańsk 2006 11

Zettabyte File System Zettabyte File System

ZFS Self Healing
ApplicationApplication Application

ZFS Mirror

ZFS Mirror

ZFS Mirror

UNIX DAYS – Gdańsk 2006

12

Zettabyte File System

Data integrity - example
# zpool create home mirror c5t40d0 c5t40d1 # cd /home/ ; ncftp sol-nv-b39-x86-dvd.iso # digest -a md5 sol-nv-b39-x86-dvd.iso a09a7af5ad25a980fa907b552a5bd523 # dd if=/dev/zero of=/dev/rdsk/c5t40d0s0 bs=1024k # digest -a md5 sol-nv-b39-x86-dvd.isoa09a7af5ad25a980fa907b552a5bd523

UNIX DAYS – Gdańsk 2006

13

Zettabyte File System

Ditto blocks
● ● ● ●

Can't read data block – EIO Can't read meta-data block – part of fs is gone In most cases meta-data is less than 2% So replicate meta-data blocks
●

Additional to pool protection (mirror, raid-z)

UNIX DAYS – Gdańsk 2006

14

Zettabyte File System

Ditto blocks
3xmeta-data blocks with pool information ● 2x meta-data blocks with fs information ● 1x user data blocks
●
●

Planned are 2x and 3x user data ditto blocks per fs

●

Keep each copy on separate vdev ● Spread block with only one drive
●

With user ditto blocks you get mirroring on single disk
UNIX DAYS – Gdańsk 2006 15

Zettabyte File System

RAID-5
●

RAID-5 write hole – silent data...