VxFS

1
VxFS VxVM on Linux Giving Linux enterprise
strength

• Hans van Rietschote
• 1-650-318-4066
• hans_at_veritas.com

2
VERITAS

• Data Availability Software-only products
• gt 1B revenue in 2000
• gt 4500 people
• HQ in Silicon valley USA
• Engineering locations all over USA, Pune and
  London (Linux Kernel team)
• Always looking for Storage SW engineers
• http//www.veritas.com/jobs

3
VxFS and VxVM background

• Developed in the late 80s / early 90s
• Original platform was UNIX System V Release 4.0
• Now ported to many UNIX variants and non-UNIX
  environments (Solaris, HP-UX, NT, W2K, Linux,
  AIX)
• First choice for many customers running in
  heterogeneous environments (VxFS/VxVM are the
  same everywhere!)

4
VxFS main features(1)

• Journaling to avoid a full fsck.
• Small amounts of data can be in intent log
• Fixed variable sized Extent based allocation
• On-line grow and shrink
• on-line administration (extent directory
  re-organization, defrag)
• Snap-shots
• Clustered file-system (direct multi-host access)

5
VxFS main features(2)

• Sys Admin Control over
• allocation policies,
• mount options,
• caching advisories
• and many other I/O policies.
• Raw I/O performance through Quick I/O
• greater than raw I/O performance through cached
  Quick I/O
• Intent logs can be on separate device (extra
  speed if on solid state disks)

6
VxVM Terminology

• subdisk part of a disk
• plex logical object one or more subdisks.
• volume logical object one or more plexes.
  Accessed just as you would a normal disk or other
  device, where the data lives
• disk groups A collection of VxVM managed disks.
  can be deported from one machine and imported to
  another machine with no special handling.

7
VxVm Volume layouts(1)

• concatenated subdisks combined in a linear
  (non-contiguous) fashion. No redundancy.
• spanning a concatenation across multiple disks.
  No redundancy.
• mirror(RAID1) data is mirrored to each (up to
  32) plex. Redundancy speed
• stripe (RAID0) data is interleaved across two or
  more plexes. No redundancy, better throughput.

8
VxVm Volume layouts(2)

• RAID5 RAID5 done in software, striping with a
  parity column. Redundancy.
• mirror-stripe two or more mirrors that are
  striped. Better throughput with redundancy. As
  long as one valid plex remains, data is
  available.
• stripe-mirror each column of the stripe is
  mirrored. This allows a finer failure
  granularity than mirror-stripe.

9
VxVM Features(1)

• All VxVM objects recognized at boot
• All configuration changes are done
  transactionally, can be reversed/resumed
• Online relayout/resize
• Hot (un)relocation a spare pool of disks can be
  setup, so that if a failure occurs, the disk is
  replaced automatically.
• dirty region log for mirrors a log kept of
  regions that have been written on a mirror. This
  allows only those regions to be re-synced
  following system failures.

10
VxVM Features(2)

• RAID5 log a log of data and parity kept until
  writes actually reach the disks. Prevents
  corruption when a disk and the system fail.
• fast mirror resync provides faster
  resyncronization of mirrors through use of a log
  area.
• rootability put root disk under VxVM, so can
  be mirrored
• dynamic multi-pathing provides a single access
  point and management of multi-pathed disk arrays.

11
VxVM Features(3)

• clustered volume manager provides access
  through a cluster of machines, to shared volumes.
  As long as one node is up, data is available.
• Java-based GUI. Same GUI across all platforms.
  Server and client can run independently.

12
Linux port goals

• Only support 2.4.x due to Linux-VM/VFS
  differences (important for VxFS to emulate page
  cache style OSs)
• Retain existing platform independence (VxFS/VxVM
  has a common source base with platform dependent
  layer)
• Avoid requiring changes to the Linux kernel
  wherever possible (allows easier migration to
  future releases)

13
Linux Port status

• Started with 2.3.x in October 1999
• Currently using 2.4.0-test9
• Main implementation is complete. Both considered
  at beta stage.
• (No) changes required to base kernel
• Fragmentation problems encountered with kernel
  memory allocator and out of memory problems
  (killing processes).

14
Commercial and non-commercial do they fit?

• There is room for both. Open source solutions for
  low to middle end of the market and commercial
  solutions for the mid to high end market
• Many high-end customers asking for VERITAS
  products since they operate heterogeneous
  environments and want the same storage software
  on all machines
• Customers also demand 24x7 enterprise class
  support
• Sharing disks between different OSs

15
Roadmap going forward

• Currently awaiting 2.4 Linux kernel release
• VERITAS Linux based Storage Appliances
• Different pricing models being considered. For
  example a free "Lite" version as per other
  operating systems (e.g. HP-UX, W2K).
• Other VERITAS products currently being ported,
  NetBackup and VCS.
• Giving Linux enterprise strength