Accelerating backup by writing to performance ranked memory
One example method includes selecting a best performing memory for an operation. A data protection system may maintain a global view of performance data for multiple namespaces. The performance data may reflect latency and/or bandwidth for each of the namespaces. The global view may be updated. When performing an operation, the best performing namespace can be selected from the global view based on performance.
1. A method, comprising:
generating a heterogeneous index for each namespace of a backup storage system, wherein the storage system includes a plurality of servers and each server is associated with at least one namespace and wherein the heterogeneous index allows a destination latency of the namespace to be determined;
storing the heterogeneous index for each namespace at the associated server;
determining a fabric latency and a source latency, wherein the fabric latency relates to a latency between a data protection system and the backup storage system;
generating a global view that is stored at a data protection system, wherein the global view includes, for each namespace, an overall latency that includes the destination latency, the fabric latency, and the source latency; and
selecting a namespace from the global view for an operation based on the overall latency.
2. The method of claim 1 , further comprising deriving the destination latency from the heterogeneous index, wherein the heterogeneous index includes data from an HMAT or other ACPI table.
3. The method of claim 1 , wherein the destination latency is determined from the heterogeneous index and a component that accounts for reads and writes of the workload.
4. The method of claim 1 , further comprising:
determining the fabric latency by sending a command from the source and waiting for a reply, wherein the fabric latency includes a time from sending the command to receiving the reply; and
determining the source delay accounts based on a time related to sending a command from the source to a network card and receiving a response to the command from the network card, wherein the network card that communicates with the fabric.
5. The method of claim 4 , wherein the command is a NULL command.
6. The method of claim 1 , further comprising updating the global profile periodically or based on an event.
7. The method of claim 1 , further comprising updating the heterogeneous index of each namespace and wherein the namespace is a storage class memory.
8. The method of claim 1 , wherein selecting the namespace accounts for existing workloads, wherein the overall delay accounts for existing workloads.
9. The method of claim 1 , wherein the global view includes bandwidth information for each namespace, wherein the selection is based on bandwidth and/or the overall latency.
10. The method of claim 1 , further comprising selecting the best performing namespace within the backup storage system based on the global view for the operation.
11. A non-transitory storage medium having stored therein instructions that are executable by one or more hardware processors to perform operations comprising:
generating a heterogeneous index for each namespace of a backup storage system, wherein the storage system includes a plurality of servers and each server is associated with at least one namespace and wherein the heterogeneous index allows a destination latency of the namespace to be determined;
storing the heterogeneous index for each namespace at the associated server;
determining a fabric latency and a source latency, wherein the fabric latency relates to a latency between a data protection system and the backup storage system;
generating a global view that is stored at a data protection system, wherein the global view includes, for each namespace, an overall latency that includes the destination latency, the fabric latency, and the source latency; and
selecting a namespace from the global view for an operation based on the overall latencies.
12. The non-transitory storage medium of claim 11 , further comprising deriving the destination latency from the heterogeneous index, wherein the heterogeneous index includes data from an HMAT or other ACPI table.
13. The non-transitory storage medium of claim 11 , wherein the destination latency is determined from the heterogeneous index and a component that accounts for reads and writes of the workload.
14. The non-transitory storage medium of claim 11 , further comprising:
determining the fabric latency by sending a command from the source and waiting for a reply, wherein the fabric latency includes a time from sending the command to receiving the reply; and
determining the source delay accounts based on a time related to sending a command from the source to a network card and receiving a response to the command from the network card, wherein the network card that communicates with the fabric.
15. The non-transitory storage medium of claim 14 , wherein the command is a NULL command and wherein the namespace is a storage class memory.
16. The non-transitory storage medium of claim 11 , further comprising updating the global profile periodically or based on an event.
17. The non-transitory storage medium of claim 11 , further comprising updating the heterogeneous index of each namespace.
18. The non-transitory storage medium of claim 11 , wherein selecting the namespace accounts for existing workloads, wherein the overall delay accounts for existing workloads.
19. The non-transitory storage medium of claim 11 , wherein the global view includes bandwidth information for each namespace, wherein the selection is based on bandwidth and/or the overall latency.
20. The non-transitory storage medium of claim 11 , further comprising selecting the best performing namespace within the backup storage system based on the global view for the operation.