Research Data Backup with CrashPlan

We  started an evaluation/trial of CrashPlan Enterprise as a backup solution for research data earlier in the year.

This trial finished a few months back but I just wanted to document some key results from this work.

  • CrashPlan is a solution for personal data backup across a range of platforms – Mac, Linux and Windows. It consists of a Java based client and for the Enterprise version a Java based server product.
  • During the evaluation I believe CrashPlan performed very well in terms of installation simplicity, configurability and on-going administration. CrashPlan allows for devolution of responsibility for backups to each college with a fully developed administrative role model.
  • The trial enabled us to fully evaluate it as a potential research data backup solution for use with our EMC Atmos. In our trial, CrashPlan server software was installed on all the Atmos IFS servers with a management GUI installed on a seperate server. The Atmos IFS servers mount the Atmos object space as a file system and this allowed us to offer Atmos space to CrashPlan as a normal file system.
  • Initial results were good, backups occurred at a reasonable speed. However after a period, backup failure would start to occur. It became apparent that this is because of the requirement that a CrashPlan server must maintain its backup files periodically and that during this time a backup cannot take place. The characteristic of this maintenance is many small multiple I/O operations on the backup files which because they are actually stored on Atmos as objects, make the maintenance operation too slow to sustain a reliable backup service. The CrashPlan backup files were remaining in maintenance mode for extremely long periods (days/weeks).
  • There is no work around to this fundamental characteristic of high start-up latency for Atmos I/O.
  • Consequently I cannot recommend CrashPlan as a solution, with Atmos as the back end, for research data backup.
  • I would however highly recommend CrashPlan as a very good solution for general research data backup if we could provide a central backup storage file system with low latency I/O.



Posted under Technical development

This post was written by Peter Leggett on October 10, 2012

3 Comments so far

  1. Eric October 12, 2012 20:16

    Great, precise, review of CrashPlan. It’s interesting that because of the I/O of your application that CrashPlan isn’t a solution that works for you.

    I assume you were backing up open files (default of CrashPlan). What if you were to deselect that? Then the I/O may not be an issue.

  2. MD Sharma October 14, 2012 22:09

    An accurate review indeed. I was allowed to test drive during the trial and am in agreement with the findings.

    I have been using CrashPlan with multiple non EMC Atmos systems as the storage nodes and it is working like a charm. Would highly recommend CrashPlan for our College backup systems (if and when they exist)

  3. Peter Leggett October 16, 2012 11:32

    Eric, it’s not the I/O of our clients or our application that is the problem – it’s the CrashPlan server. We were evaluating CrashPlan Enterprise which lets you run your own CrashPlan server on your own hardware.

More Blog Post