ZEBRA, ZAMBONI, ZEN & THE ART OF ZFS Maureen Chew Staff - - PowerPoint PPT Presentation

zebra zamboni zen the art of zfs
SMART_READER_LITE
LIVE PREVIEW

ZEBRA, ZAMBONI, ZEN & THE ART OF ZFS Maureen Chew Staff - - PowerPoint PPT Presentation

ZEBRA, ZAMBONI, ZEN & THE ART OF ZFS Maureen Chew Staff Engineer Sun Microsystems maureen.chew@sun.com SAS Global Forum 2007, Orlando, FL, paper 198-2007 Agenda Whats with all the ZZZZs? Zebra Zamboni > Data Integrity


slide-1
SLIDE 1

ZEBRA, ZAMBONI, ZEN & THE ART OF ZFS

Maureen Chew

Staff Engineer Sun Microsystems maureen.chew@sun.com

SAS Global Forum 2007, Orlando, FL, paper 198-2007

slide-2
SLIDE 2

SAS Global Forum 2007, Orlando, FL, paper 198-2007

Agenda

  • What’s with all the ZZZZ’s?
  • Zebra
  • Zamboni

> Data Integrity > Reliability > Performance

  • Case Study – Solaris ZFS & Containers in use at

the CDC

  • Zen
slide-3
SLIDE 3

SAS Global Forum 2007, Orlando, FL, paper 198-2007

What's with all the ZZZZ's?

  • Zebra – known for its stripes, ZFS is a volume

manager and includes striping capabilities

  • Zamboni – frees the rink of all the bumps, lumps

and other irregularities that can cause disastrous crashes in the data management arena

  • Zen – ...enlightenment upon means of direct &

intuitive insight.....

  • Solaris 10 ZFS brings the Zen to IT data

management

slide-4
SLIDE 4

SAS Global Forum 2007, Orlando, FL, paper 198-2007

Agenda

  • What’s with all the ZZZZ’s?
  • Zebra
  • Zamboni

> Data Integrity > Reliability > Performance

  • Case Study – Solaris ZFS & Containers in use at

the CDC

slide-5
SLIDE 5

SAS Global Forum 2007, Orlando, FL, paper 198-2007

Zebra stripes

  • Raid 0, 1, Z (like Raid 5)
  • Easy combinations

> Mirrored stripes > Concatenated RaidZ virtual

devices

slide-6
SLIDE 6
slide-7
SLIDE 7

SAS Global Forum 2007, Orlando, FL, paper 198-2007

Zebra Stripes

  • Can create 1 or more file systems in a pool
  • File systems

> As easy as a directory > Can have individual quotas > Can be snapshot'ed, cloned > Can be archived/re-created > Share the same pool of storage resources which can

accommodate dynamic changes in size without distruption to service.

slide-8
SLIDE 8

SAS Global Forum 2007, Orlando, FL, paper 198-2007

Agenda

  • What’s with all the ZZZZ’s?
  • Zebra
  • Zamboni

> Data Integrity > Reliability > Performance

  • Case Study – Solaris ZFS & Containers in use at

the CDC

slide-9
SLIDE 9

SAS Global Forum 2007, Orlando, FL, paper 198-2007

Zamboni – Data Integrity

  • All data protected by 256 bit checksums
  • Data consistency maintained at all times, no need

for fsck(1M)

> Data written to new disk blocks prior to pointers being

updated.

> No need for fsck

  • Zpool scrub
slide-10
SLIDE 10
slide-11
SLIDE 11

SAS Global Forum 2007, Orlando, FL, paper 198-2007

ZFS Reliability – Try this at Home!

  • Geek (ie: cheap &

weird) redundant storage device

  • 3 USB hubs, 12

memory sticks

  • 1 command, voila, a

highly available, ready- to-use ZFS file system

10 minute humor alert: Check out the virtues of ZFS at: http://blogs.sun.com/constantin/entry/csi_munich_how_to_save

slide-12
SLIDE 12

SAS Global Forum 2007, Orlando, FL, paper 198-2007

ZFS Reliability – RAID-Z at Work!

  • 4 sticks removed
  • 1/3 of storage vanished
  • Volume takes a licking,

but keeps on ticking!

slide-13
SLIDE 13

SAS Global Forum 2007, Orlando, FL, paper 198-2007

ZFS Reliability – RAID-Z at Work!

scrub: none requested config: NAME STATE READ WRITE CKSUM saspool ONLINE pool: saspool state: ONLINE scrub: none requested config: NAME STATE READ WRITE CKSUM saspool ONLINE 0 0 0 raidz1 DEGRADED 0 56 0 cannot open c9t0d0p0 ONLINE 0 0 0 c3t0d0p0 ONLINE 0 0 0 c8t0d0p0 ONLINE 0 0 0 raidz1 DEGRADED 0 26 0 cannot open c13t0d0p0 ONLINE 0 0 0 c7t0d0p0 ONLINE 0 0 0 c10t0d0p0 ONLINE 0 0 0 raidz1 DEGRADED 0 4 0 cannot open c6t0d0p0 ONLINE 0 0 0 c15t0d0p0 ONLINE 0 0 0 c12t0d0p0 ONLINE 0 0 0 raidz1 DEGRADED 0 52 0 cannot open c18t0d0p0 ONLINE 0 0 0 c5t0d0p0 ONLINE 0 0 0 c11t0d0p0 ONLINE 0 0 0 errors: No known data errors

Zpool(1) status shows degraded volume, but no data errors

slide-14
SLIDE 14
slide-15
SLIDE 15

SAS Global Forum 2007, Orlando, FL, paper 198-2007

Performance

  • A dozen SAS programs

run on Sun Fire X4600

  • ZFS vs. UFS, Linux

EXT3 on similar system

  • ZFS performance*

> ZFS: ~3.5 hrs > UFS: ~4.0 hrs > Linux EXT3: ~4.5 hrs

00:00:00 00:36:00 01:12:00 01:48:00 02:24:00 03:00:00 03:36:00 04:12:00 04:48:00 05:24:00

Cumulative Performance Results

UFS ZFS EXT3

Solaris 10 UFS/ZFS, Linux EXT3 Total Time

* your mileage may vary

slide-16
SLIDE 16

SAS Global Forum 2007, Orlando, FL, paper 198-2007

Agenda

  • What’s with all the ZZZZ’s?
  • Zebra
  • Zamboni

> Data Integrity > Reliability > Performance

  • Case Study – Solaris ZFS & Containers in use at

the CDC

slide-17
SLIDE 17

SAS Global Forum 2007, Orlando, FL, paper 198-2007

Case Study

  • World Health Concerns at the Center for Disease

Control – CDC

  • Solaris 10 – ZFS & Containers in Action
  • Anantha Srirama, Unisys/Nidhi Consulting,

BioSense Infrastructure Architect faced daunting challenges...

slide-18
SLIDE 18

SAS Global Forum 2007, Orlando, FL, paper 198-2007

Challenges

  • Nightly update from SQL server to SAS SPD Server
  • SPD Server not available during update phase
  • Nightly maintenance window for updates shrinks as

data volume grows

  • SPD Server data replication

> Cannot have duplicate SPD Server domains on same

system; solution would be:

> Costly > Complicated > Time consuming

slide-19
SLIDE 19

SAS Global Forum 2007, Orlando, FL, paper 198-2007

Solution – Simple and Ingenious Solaris 10 Containers and Clones

SPD Server Data Read/Write ZFS Filesystem SPD Server Data Read/Write ZFS Filesystem (Virtual) Read/Write Read

E2900 running Solaris 10 Update 3 Global Zone

SunFire E2900

S un

Local Zone (Virtual)

SunFire E2900

S un

slide-20
SLIDE 20

SAS Global Forum 2007, Orlando, FL, paper 198-2007

Solution – Simple and Ingenious!

  • Clone the SPD Server file systems
  • Marry the clone to a Solaris Container
  • Create appropriate SAS metadata librefs/access

templates for:

> Live SPD Server in global zone which only a few SAS

programmers can access/update

> Cloned SPD Server is Container which all authorized

users can access

  • Happy users, admins happy, updates can take as

long as needed

slide-21
SLIDE 21

SAS Global Forum 2007, Orlando, FL, paper 198-2007

Solution – Simple and Ingenious Solaris 10 Containers and Clones

SPD Server Data Read/Write ZFS Filesystem SPD Server Data Read/Write ZFS Filesystem (Virtual) Read/Write Read

E2900 running Solaris 10 Update 3 Global Zone

SunFire E2900

S un

Local Zone (Virtual)

SunFire E2900

S un

1

Daily update jobs complete

2

Shutdown Zone

3

ZFS Clone Destroy Create

4

Startup Zone

slide-22
SLIDE 22

SAS Global Forum 2007, Orlando, FL, paper 198-2007

Where's the WOW WOW? ?

  • Concept to deployment – 4 hours
  • No incremental cost
  • Happy users, IT management, system admins

> Oxymoron

  • Instead of “Sleepless in Seattle”,

Anantha is “Asleep in Atlanta”

> No worries at night that the update won't be finished in

time for the next morning

slide-23
SLIDE 23
slide-24
SLIDE 24

SAS Global Forum 2007, Orlando, FL, paper 198-2007

Zen

  • ... enlightement but means of direct and intuitive

insight

> Cost > Flexibility > Reliability > Lack of Complexity > High Performance

Solaris ZFS

slide-25
SLIDE 25

ZEBRA, ZAMBONI, ZEN & THE ART OF ZFS

Maureen Chew

Staff Engineer Sun Microsystems maureen.chew@sun.com

SAS Global Forum 2007, Orlando, FL, paper 198-2007