Academia.eduAcademia.edu

Figure 1 – uploaded by Georgiana Vasile

See full PDFdownloadDownload figure
Protector: It is in charge of requesting the observers to perform checkpoints and storing the checkpoint files in its own Stable Storage (SS). It is also in charge of the detection of failures by verifying that the node that it’s protecting is working and, should if fail, it performs the restoration of the process that failed by launching the latest checkpoint.

Figure 1 Protector: It is in charge of requesting the observers to perform checkpoints and storing the checkpoint files in its own Stable Storage (SS). It is also in charge of the detection of failures by verifying that the node that it’s protecting is working and, should if fail, it performs the restoration of the process that failed by launching the latest checkpoint.

Related Figures (8)

Besides the regular RADIC architecture designed to protect from node fails, H-RADIC will protect applications from crashing in the event of multiple fails, granting that the application finishes its execution despite the fails. When the virtual nodes fail, the physical nodes that the virtual ones are mounted on fail and/or whenever there is loss of Recovery: Protector Ta, _ restarts/rolls-back the processes running in Ng, using the data saved in the SS from the last checkpoint, if the system has a spare node (Nz_), the processes will be restarted in it, otherwise the processes are restarted in the node
Besides the regular RADIC architecture designed to protect from node fails, H-RADIC will protect applications from crashing in the event of multiple fails, granting that the application finishes its execution despite the fails. When the virtual nodes fail, the physical nodes that the virtual ones are mounted on fail and/or whenever there is loss of Recovery: Protector Ta, _ restarts/rolls-back the processes running in Ng, using the data saved in the SS from the last checkpoint, if the system has a spare node (Nz_), the processes will be restarted in it, otherwise the processes are restarted in the node
The main difference between the two restart options is the way to store the checkpoints. In the case that there is a spare cluster available in another cloud (Figure 3), H-RADIC will be working as usual, but if there are no spare nodes available in a third cloud, the checkpoint will be restarted in the spare nodes of the cloud that has the checkpoint. After the restart, the two clusters (X and Y’ in Figure 4) will send their checkpoints to Cluster Z and vice versa. Then the checkpoint files are sent to the nodes that are in the spare cluster.
The main difference between the two restart options is the way to store the checkpoints. In the case that there is a spare cluster available in another cloud (Figure 3), H-RADIC will be working as usual, but if there are no spare nodes available in a third cloud, the checkpoint will be restarted in the spare nodes of the cloud that has the checkpoint. After the restart, the two clusters (X and Y’ in Figure 4) will send their checkpoints to Cluster Z and vice versa. Then the checkpoint files are sent to the nodes that are in the spare cluster.
Finally, the error masking function pseudocode is described in Algorithm 4.
Finally, the error masking function pseudocode is described in Algorithm 4.
Figure 5 - H-RADIC Recovery function - spare nodes/cluster in the same cloud and not more clouds left. controller, the storage and the physical computer that the virtual nodes are mounted on. These vulnerabilities will now allow us to be able to guaranty availability.
Figure 5 - H-RADIC Recovery function - spare nodes/cluster in the same cloud and not more clouds left. controller, the storage and the physical computer that the virtual nodes are mounted on. These vulnerabilities will now allow us to be able to guaranty availability.
Another experiment was performed where an error was induced around the middle of the application’s execution and the we took the time that the application took to restart form the latest checkpoint and the time that the application took to restart from the beginning, as shown in Figure 7.
Another experiment was performed where an error was induced around the middle of the application’s execution and the we took the time that the application took to restart form the latest checkpoint and the time that the application took to restart from the beginning, as shown in Figure 7.
Figure 6 - H-RADIC percentage of time overhead. an induced error. Then by taking the increase percentage of time in 2) and 3), we attain Figure 6.
Figure 6 - H-RADIC percentage of time overhead. an induced error. Then by taking the increase percentage of time in 2) and 3), we attain Figure 6.
Table 2 - Variables description. 5. Conclusion and future work
Table 2 - Variables description. 5. Conclusion and future work
Related topics:
Computer ScienceComputer Science and Information Technology
580 California St., Suite 400
San Francisco, CA, 94104
© 2025 Academia. All rights reserved