Optimized distributed deduplication for distributed cluster
Distributed deduplication wherein runtime performance of dedup pipelines in all nodes is monitored. The bottleneck for each pipeline is identified and machine resources from different nodes are reallocated to seek to balance the costs of each stage of each task in each of the pipelines. While the overall cost for each task may remain the same, stalls may be eliminated such that the total cost to complete all the tasks is reduced. The global dedup ratio and the local compression ratio may be used to weight certain stage costs.
1. A computer-implemented method for performing distributed deduplication in a cluster of machines forming a plurality of nodes, comprising:
dividing a deduplication process into a plurality of process stages, wherein a group of stages form a task to be executed as a pipeline;
assigning selected nodes from the plurality of nodes to perform assigned process stages and tasks, wherein each of the selected nodes services a plurality of the pipelines;
within each node of the cluster, assigning computing and storage resources to perform each stage of a deduplication task;
while the deduplication process is performed within the cluster, for each pipeline periodically performing:
collecting performance data of each of the computing and storage resources and using the performance data to determine process cost for each of the stages of tasks within a corresponding pipeline;
for each task reassigning, at least partially, computing and storage resources from stages having lower process cost to stages having higher process cost;
updating resources utilization of each node within the cluster.
2. The method of claim 1 , wherein reassigning, at least partially, computing and storage resources includes reassigning computing and storage resources of one node to perform stages of tasks on another node.
3. The method of claim 1 , wherein reassigning, at least partially, computing and storage resources is calculated to balance the process cost of the stages of each task.
4. The method of claim 3 , further comprising calculating a global deduplication ratio and wherein determining process cost for each of the stages comprises weighting the process cost of at least one stage by the global deduplication ratio.
5. The method of claim 4 , further comprising calculating a local compression ratio and wherein determining process cost for each of the stages comprises weighting the process cost of at least one stage by the local compression ratio.
6. The method of claim 1 , wherein the plurality of pipelines includes at least a deduplication pipeline, a restore pipeline, and a garbage collection pipeline.
7. The method of claim 1 , further comprising at each node maintaining information of computing and storage resources of other nodes within the cluster and sending local information of computing and storage resources to other nodes within the cluster.
8. A non-transitory computer-readable medium programmed with executable instructions that, when executed by a processing system having at least one hardware processor, perform operations for performing distributed deduplication in a cluster of machines forming a plurality of nodes, the operations comprising:
dividing a deduplication process into a plurality of process stages, wherein a group of stages form a task to be executed as a pipeline;
assigning selected nodes from the plurality of nodes to perform assigned process stages and tasks, wherein each of the selected nodes services a plurality of the pipelines;
within each node of the cluster, assigning computing and storage resources to perform each stage of a deduplication task;
while the deduplication process is performed within the cluster, for each pipeline periodically performing:
collecting performance data of each of the computing and storage resources and using the performance data to determine process cost for each of the stages of tasks within a corresponding pipeline;
for each task reassigning, at least partially, computing and storage resources from stages having lower process cost to stages having higher process cost;
updating resources utilization of each node within the cluster.
9. The medium of claim 8 , wherein reassigning, at least partially, computing and storage resources includes reassigning computing and storage resources of one node to perform stages of tasks on another node.
10. The medium of claim 8 , wherein reassigning, at least partially, computing and storage resources is calculated to balance the process cost of the stages of each task.
11. The medium of claim 10 , wherein the operations further comprise calculating a global deduplication ratio and wherein determining process cost for each of the stages comprises weighting the process cost of at least one stage by the global deduplication ratio.
12. The medium of claim 11 , wherein the operations further comprise calculating a local compression ratio and wherein determining process cost for each of the stages comprises weighting the process cost of at least one stage by the local compression ratio.
13. The medium of claim 8 , wherein the plurality of pipelines includes at least a deduplication pipeline, a restore pipeline, and a garbage collection pipeline.
14. The medium of claim 8 , wherein the operations further comprise at each node maintaining information of computing and storage resources of other nodes within the cluster and sending local information of computing and storage resources to other nodes within the cluster.
15. A system comprising:
a processing system having at least one hardware processor, the processing system coupled to a memory programmed with executable instructions that, when executed by the processing system, perform operations for performing distributed deduplication in a cluster of machines forming a plurality of nodes, the operations comprising:
dividing a deduplication process into a plurality of process stages, wherein a group of stages form a task to be executed as a pipeline;
assigning selected nodes from the plurality of nodes to perform assigned process stages and tasks, wherein each of the selected nodes services a plurality of the pipelines;
within each node of the cluster, assigning computing and storage resources to perform each stage of a deduplication task;
while the deduplication process is performed within the cluster, for each pipeline periodically performing:
collecting performance data of each of the computing and storage resources and using the performance data to determine process cost for each of the stages of tasks within a corresponding pipeline;
for each task reassigning, at least partially, computing and storage resources from stages having lower process cost to stages having higher process cost;
updating resources utilization of each node within the cluster.
16. The system of claim 15 , wherein reassigning, at least partially, computing and storage resources includes reassigning computing and storage resources of one node to perform stages of tasks on another node.
17. The system of claim 15 , wherein reassigning, at least partially, computing and storage resources is calculated to balance the process cost of the stages of each task.
18. The system of claim 17 , wherein the operations further comprise calculating a global deduplication ratio and wherein determining process cost for each of the stages comprises weighting the process cost of at least one stage by the global deduplication ratio.
19. The medium of claim 18 , wherein the operations further comprise calculating a local compression ratio and wherein determining process cost for each of the stages comprises weighting the process cost of at least one stage by the local compression ratio.
20. The medium of claim 15 , wherein the plurality of pipelines includes at least a deduplication pipeline, a restore pipeline, and a garbage collection pipeline.
21. The medium of claim 15 , wherein the operations further comprise at each node maintaining information of computing and storage resources of other nodes within the cluster and sending local information of computing and storage resources to other nodes within the cluster.