Docstoc

Westgrid UBC TRIUMF Site

Document Sample
Westgrid UBC TRIUMF Site Powered By Docstoc
					       HEPiX Spring 2006



AMANDA @ TRIUMF
Steven McDonald & Konstantin Olchanski
TRIUMF Network & Computing Services
       steven.mcdonald@triumf.ca




            HEPiX, CASPUR, April 3-7, 2006 – Steve McDonald
                        Perspective
• Amanda is not in the same class of TSM
   – Make no mistake this is *not* a TRIUMF T1 storage solution

• Not something for large experimental data sets
   – Requiring frequent access


• More suitable for group, department sys admins for backing-up
  user desktops and servers,
   – Infrequent data recovery/access
   – Disaster recovery
   – what it was designed for

• Low cost, easy setup, low maintenance backup of desktops and
  servers

                            HEPiX, CASPUR, April 3-7, 2006 – Steve McDonald
          What’s on the Menu

•   What type of backups?
•   Hardware
•   Software configuration
•   Our environment
•   Typical nightly usage
•   Performance
•   Advantages & disadvantages
•   Conclusions

                    HEPiX, CASPUR, April 3-7, 2006 – Steve McDonald
    Background Requirements
• What are we attempting to do?
   – Archive data?
   – Disaster recovery?
   – Careless file deletion and recovery?

• Would like something to do all the above with
   – Low cost, economically
   – Minimal man-power resources, monitoring, maintenance

• Amanda can do this for us with some compromise

• Obviously this is NOT a T1 Backup solution
                      HEPiX, CASPUR, April 3-7, 2006 – Steve McDonald
                      Amanda
    Advanced Maryland Automatic Network Disk Archiver
          opensource from http://www.amanda.org

•    Amadmin                         •    Amoverview
•    Amcheck                         •    Amplot
•    Amcleanup                       •    Amrecover
•    Amadd                           •    Amreport
•    Amdump                          •    Amrestore
•    Amflush                         •    Amrmtape
•    Amgetconf                       •    Amstatus
•    Amlabel                         •    Amtape
•    Ammnt                           •    Amtapetype
•    Amverify                        •    Amtoc


                      HEPiX, CASPUR, April 3-7, 2006 – Steve McDonald
      Supported Operating Systems
•   Linux (~150)                  •    SCO
•   Solaris (~6)                  •    FreeBSD
•   Ultrix                        •    AIX
•   Digital Unix                  •    Mac OS X
•   SunOS 4                       •    Microsoft Windows ?
•   HPUX
•   Sinix
•   IRIX

                   HEPiX, CASPUR, April 3-7, 2006 – Steve McDonald
                         Amanda Hardware
• Amanda system (~$10k US)
   – SL4.2
   – Dual Opteron 248 2.2 GHz
      • 3U/Dual power
      • 2G Memory
      • Data Storage 16 x400G WD disks ~ 6TB
      • System disks 2x 80G

• Overland Neo 2000 tape library
    – 26 slot capacity

• SDLT 600 drive
    – SDLT2 300GB native per tape

• ~7.5TB native DLT2 tape capacity

• Library & DLT600 drive ~$12k US
                                    HEPiX, CASPUR, April 3-7, 2006 – Steve McDonald
      Disk I/O Configuration

• System disks 2x 80GB software RAID1 mirror

• Backup Storage 16x 400GB software RAID5
  – ~6TB useable space

• Disk I/O performance from Bonnie++
  – 200 MB/sec block writes
  – 280 MB/sec block reads

                  HEPiX, CASPUR, April 3-7, 2006 – Steve McDonald
            What is backed up
• ~ 150 machines
  – Mostly linux RH7.3 – SL4.2
  – Some Solaris

• Servers
  – Everything including system and home directories

• Client machines
  – /etc ( ~20MB )
  – /home (no enforced limits, but not data files)
     • 10-60GB is typical




                       HEPiX, CASPUR, April 3-7, 2006 – Steve McDonald
     Amanda Client & Server Configuration

• On the client:
   – edit ~amanda/.amandahosts, add line "amanda.triumf.ca amanda"
   – edit /etc/hosts.allow, add line "amandad: amanda.triumf.ca"
   – chkconfig amanda on



• On the server:
   – edit ~amanda/daily/disklist, add entry "client /home1 comp-usr
     #clientusername@triumf.ca"
   – run "amcheck -c daily client"

                          HEPiX, CASPUR, April 3-7, 2006 – Steve McDonald
                                Some statistical distributions

                       Distribution of filesystem sizes                                          Time for full backups
                                                          1-2MB

                  45                                      2-5MB                                                          0-5 sec
                                                                                            60
                                                          5-10MB                                                         5-10 sec
                  40
                                                          10-20MB                           50                           10-30 sec
                  35                                      20-50MB                                                        30-60sec
No. Filesystems




                                                                           No Filesystems
                  30                                      50-100MB                          40                           1-5 mins
                  25                                      100-200MB                                                      10-20 mins
                                       vb                 200-500MB                         30                           20-40 mins
                  20
                                                          500-1000MB                                                     40-60 mins
                  15                                                                        20
                                                          1-2GB                                                          1-2 hrs
                  10                                      2-5GB                                                          2-3 hrs
                   5                                      5-10GB                            10                           3-4 hrs
                                                          10-50GB                                                        4-5 hrs
                   0
                                                                                            0                            5-6 hrs
                                  1                       50-100GB
                                                                                                       1
                                Size                      100-200GB                                                      6-7 hrs
                                                                                                     Time
                                                          >200GB                                                         > 7 hrs




                                                          HEPiX, CASPUR, April 3-7, 2006 – Steve McDonald
How do we use Amanda




      HEPiX, CASPUR, April 3-7, 2006 – Steve McDonald
         Amanda Network Usage

Daily




Weekly


               HEPiX, CASPUR, April 3-7, 2006 – Steve McDonald
Restoring Data (1)




     HEPiX, CASPUR, April 3-7, 2006 – Steve McDonald
Restoring Data (2)




     HEPiX, CASPUR, April 3-7, 2006 – Steve McDonald
                       Amanda
• Advantages                         • Disadvantages
  – Low co$t (free)                        – No nice Web/GUI
  – Pre-installed on RH,SL,FC                Administrative Interface
  – Not resource intensive
  – Highly configurable                    – Had to write scripts to
  – Full featured                            visualize summary data
  – Runs on wide range of
    hardware                               – No long term data trending
  – Low maintenance
  – CLI highly informative                 – Difficult to archive data
  – Easy restore operation                   unless you use lot$ of tapes


                       HEPiX, CASPUR, April 3-7, 2006 – Steve McDonald
                    Conclusion
• Overall Amanda provides a very reliable and configurable
  backup mechanism for our particular environment at
  TRIUMF

• At a low cost both in
   – Price
   – Man power upkeep and maintenance

• Lots of room for growth.

• More work on importing data into a Database would be
  nice for long term trending and statistical analysis … on
  my todo …

                       HEPiX, CASPUR, April 3-7, 2006 – Steve McDonald

				
DOCUMENT INFO
Shared By:
Categories:
Tags:
Stats:
views:6
posted:4/10/2012
language:Latin
pages:17