Rapid Recovery

after server crashes


Actually (i'm using last release when a server (agent installed) crashes or after an unpredictable reboot a base image is done.

This is really a problem on core replicated with a poor network bandwidth, because it takes several hours to replicate to the main core.


Is there something expected to avoid this behaviour on the next releases ?





  • Hi t.loudec:

    Sorry for the issues you are facing.

    Taking base images after a crash or a reboot that does not follow a gracious shutdown is, unfortunately, the expected behavior. This is due to the fact that the RapidRecovery filter driver responsible for intercepting the disk writes may cease working before the disk activity is suspended and by consequence some block changes (at times many gigabytes in size) may not be recorded. In turn, this may affect the integrity of the future recovery chains.

    Although this behavior cannot and should not be altered, there are some important upcoming features that would mitigate the issue. Among these the one that may be of greatest interest for your situation is agent size deduplication. This would diminish considerably the size of base images taken after unfortunate events as only the blocks that are not hosted in the repository will be sent through the wire, reducing the overall data transfers during new base images. 

  • Check out dedup cache size on the Core. Below is an old thread from 5.x but it covers a lot fo great information about this setting and how it works.


    Its not a great options, but its the only one available atm to address this problem
  • Hi Tudor,

    Many thanks for your help.

    When you said "there are some important upcoming features " do you have an idea when those features will be implemented ?

    Best regards

  • Tudor. I know these features are still in Dev and not set in stone. But do you have any details on how the agent side deduplication will function? Will it be a physical memory based setup like the Core? How about the sizing?
  • Sorry -- I did not play with the feature as yet so I have no cool stuff to share in this respect. I suspect that it is something like the "read-match-write" process.