The HPCC cluster's file server sharptail.wesleyan.edu serves out all home directories to all nodes at location /home. It is 10 TB in size and it currently takes the nightly process of backup 1-2 hours to churn through. Making it larger would thus generate more traffic on /home. So we've, for now while it works for us, come up with this policy:

• All users are under quota which automatically is increased by 100 GB increments.
• When a user consumes 1024 GB (1 TB) the automatic increases stop.
  • this home file system is twice a month backed up from sharptail's to greentail's disk array
  • nightly snapshots (point in time backups) are done on sharptail's disk array and stored there too

At this point users need to off load static content to other locations. Contents like old analyses, results of published papers, etc. Users typically have one local option available:

• Keep contents out of /home and migrate it to /archives (7 TB, accessible on all “tail” nodes)
  • request a directory for you in this file system and move contents to it
  • this archive file system is twice a month backed up from sharptail's to greentail's disk array
• Users with home directories of 500+ GB in size should start considering moving data to /archives

Users whom are considered inactive have their home directories relocated to /archives/inactive

• these accounts are kept around until we do an account edit and purge (has never happened so far)

The remote storage option, if your storage needs cannot be supported by /archives, is off-cluster storage. Rstore is our latest storage solution for groups and labs with such needs.

• ask your lead faculty member if your lab/group has such an area or request one
• then move your static content permanently off the HPCC cluster environment
• details can be found at RSTORE FAQ

Our file server is named sharptail.wesleyan.edu (or sharptail when on cluster) and it is a 4U integrated storage and server module with an 48TB of disk array. Moving content can severely crippled this server. /home is served out by this server to all nodes and if the server can not handle all read/write requests everything comes to a halt. So when moving content please monitor and also observe if others are currently doing something along this line. Here are some tips.

Do not use any type of copy tool with a GUI or cp/scp or s/ftp. Especially the GUI (drag&drop) are Verboten! These tools are not smart enough and frequently generated blocked processes that halt everything. Use rsync in a linux/unix environment.

Check it out:

• ssh sharptail.wesleyan.edu
• is the server busy (uptime loads < 8 are ok)
• is there memory available (free -m look at free values)
• is anybody else using rsync (ps -efl | grep rsync)
• is the server busy writing (iotop look at the M/s disk writes(q to quit), values >100-200 M/s == busy!)

Three scenarios are depicted below. When crossing the vertical boundaries you are not dealing with local content anymore, thus the content needs to flow over the network. rsync has many features, one of the important one is the use of a remote shell allowing an elegant way to cross these boundaries.

                        |         /home         |    group share     |    some lab location     
some lab location       |                       |                    |
                  <-----------> sharptail <-----------> Rstore <----------->                                
some other college      |                       |                    |    
                        |         /archives     |     lab share      |    some other college               

Some feature examples

• preserve permissions, do a checksum between source/target files, observe what will happen
  • rsync -vac –dry-run
• delete files on destination not present on source (careful!)
  • rsync –delete
• throttle the rate of traffic generated, make your sysadmin happy, use <5000
  • rsync –bwlimit=2500
• do not look inside files
  • rsync –whole-files
• use a remote shell from host to host (crossing those vertical boundaries above)
  • rsync -vac /home/my/stuff/ user@somehost.wesleyan.edu:/home/my/stuff/

Note the use of trailing slashes, it means update everything inside source stuff/ within target stuff/. If you left the first trailing slash off the above command it means put source directory stuff/ inside target directory stuff/ meaning you'll end up with target /home/my/stuff/stuff. You've been warned. Use the dry run option if unsure what will happen.

Putting it all together

# copy the dir stuff from lab or remote college to my home on HPCC in tmp area 
# (first log in to remote location)

rsync -vac --bwlimit=2500 --whole-files /home/user/stuff user@sharptail.wesleyan.edu:/home/user/tmp/

# sync my HPCC dir stuff folder into /archives locally on sharptail, then clean up
# (first log in to sharptail) 

rsync -vac --bwlimit=2500 /home/user/stuff/  /archives/user/stuff/
rm -rf /home/user/stuff/*

# generate a copy of content on Rstore disk array outside of HPCC but within wesleyan.edu
# (get paths and share names from faculty member, on sharptail do)

rsync -vac --bwlimit=2500 /home/user/stuff  user@rstoresrv0.wesleyan.edu:/data/2/labcontent/projects/

# you can also do this in reverse, log in to sharptail first

rsync -vac --bwlimt=2500 user@rstoresrv0.wesleyan.edu:/data/2/labcontent/projects/stuff  /home/user/ 

Back