Automation and optimization of data recovery after a ransomware attack
In the face of ransomware attacks, which can be increasingly difficult to effectively prevent, a solution can be considered to be the minimization of the cost and time taken to recover data and, hence business activities. Embodiments perform a restore operation that include automatically identifying the most recent healthy backup, from which data should be restored, and the prioritizing of the order in which data should be restored.
1. A method for performing a restore operation, the method comprising:
identifying a plurality of candidate backups from which to restore production data in a computing system that has experienced an event;
extracting features from the plurality of candidate backups;
scoring each of the plurality of candidate backups based at least on the extracted features;
identifying, by a backup server, a healthy backup from the plurality of candidate backups based on the scores of the plurality of backup candidates;
restoring the production data from the identified healthy backup.
2. The method of claim 1 , wherein extracting features includes extracting one or more of data extensions, data metadata and data timestamps.
3. The method of claim 1 , wherein a candidate backup is deemed to be infected when the extensions include extensions indicative of an infection or when a rate of change of the timestamps is greater than a threshold value.
4. The method of claim 1 , further comprising mining the plurality of candidate backups to extract patterns indicative of an infection.
5. The method of claim 4 , wherein the patterns indicative of infection include file change rates across backups that exceed a threshold and/or deviations in the extracted features from a historical norm.
6. The method of claim 1 , further comprising performing data augmentation with sources of information when identifying the healthy backup.
7. The method of claim 6 , wherein the sources of information include a historical database, a model of normal client behavior, and data from infected peer systems.
8. The method of claim 7 , further comprising generating the historical database or the model of normal client behavior with machine learning algorithms that have knowledge of typical behavior patterns.
9. The method of claim 7 , wherein the model of normal client behavior is based on deltas between successive backups.
10. The method of claim 6 , further comprising selecting the healthy backup based on the extracted features and the data augmentation.
11. A method for performing a restore operation, the method comprising:
determining that an event has occurred in a computing system that includes production data, wherein a plurality of candidate backups are available to the computing system such that the production data can be restored;
identifying a healthy backup from the plurality of candidate backups by:
extracting features from the plurality of candidates;
augmenting the process of identifying the healthy backup with augmented data;
evaluating the extracted features and the candidate backups based on the augmented data; and
assigning a score to each of the candidate backups based on the extracted features and the augmented data;
selecting, as the healthy backup, the candidate backup with the best score; and
restoring the production data from the selected healthy backup.
12. The method of claim 11 , wherein extracting features includes extracting one or more of data extensions, data metadata, filenames, and data timestamps.
13. The method of claim 12 , further determining whether the data extensions, data metadata, filenames and timestamps indicate an infection.
14. The method of claim 13 , further comprising comparing the data extensions from multiple backups, comparing the metadata from the multiple backups, comparing the filenames from the multiple backups, and comparing the timestamps from the multiple backups.
15. The method of claim 14 , further comprising determining rates of change from the timestamps.
16. The method of claim 11 , further comprising performing data augmentation with sources of information when identifying the healthy backup.
17. The method of claim 16 , wherein the sources of information include a historical database, a model of normal client behavior, and data from infected peer systems, wherein the model of normal client behavior is based on deltas between successive backups.
18. The method of claim 17 , further comprising generating the historical database or the model of normal client behavior with machine learning algorithms that have knowledge of typical behavior patterns.
19. The method of claim 16 , further comprising selecting the healthy backup based on the extracted features and the data augmentation.
20. A non-transitory computer readable medium comprising computer executable instructions configured for execution by a processor and configured to implement a method for performing a restore operation, the method comprising:
determining that an event has occurred in a computing system that includes production data, wherein a plurality of candidate backups are available to the computing system such that the production data can be restored;
identifying a healthy backup from the plurality of candidate backups by:
extracting features from the plurality of candidates;
augmenting the process of identifying the healthy backup with augmented data;
evaluating the extracted features and the candidate backups based on the augmented data; and
assigning a score to each of the candidate backups based on the extracted features and the augmented data;
selecting, as the healthy backup, the candidate backup with the best score; and
restoring the production data from the selected healthy backup.