When you use the extent-based backups on NAS file servers, a considerable increase in throughput for extent-based operations over non extent-based operations is seen.
Data set: 1 file of size 253GB
CIFS configuration: Two nodes, each node with the following configuration:
-
16 logical processors
-
48 GB RAM
-
Three locally attached SATA 7200 RPM disks configured in a Windows storage pool with four Commvault library mount points.
NFS configuration: Two nodes, each node with the following configuration:
-
8 logical processors
-
16 GB RAM
-
Four Commvault library mount points, one each on a locally attached SAS 7200 RAM disk, a locally attached SAS 15K RPM disk, and two iSCSI LUNs.
Number of streams used for a regular(non-extent-based) backup operation: The file is backed up using a single stream.
Number of streams used for an extent-based backup: Two nodes with the default, four-streams-per-node configuration, a total of eight streams are used to back up the file.
Number of streams used for a restore operation: Restores from non-extent-based backups use one stream for the restore operation. For restores from extent-based backups, two nodes with the default, ten streams, are used to restore the file.
The following table shows a considerable increase in throughput for extent-based operations over non extent-based operations.
|
NetApp |
Nutanix |
Qumulo |
Isilon |
|
|---|---|---|---|---|
|
CIFS backup |
117% |
127% |
122% |
68% |
|
CIFS restore |
29% |
61% |
38% |
26% |
|
NFS backup |
80% |
113% |
76% |
99% |
|
NFS restore |
46% |
70% |
37% |
63% |
Recommendations for a higher throughput
-
Overall throughput depends on many factors such as read speed from the file server, media-write speed, network links connecting the file server, proxy computer, and MediaAgent computer.
-
To mitigate the number of hops for the data by eliminating network traffic between the proxy computer and the MediaAgent computer, configure the proxy computer as the MediaAgent computer and ensure the backups are LAN-free media writes.
-
To ensure that each backup stream is independent, ensure that enough library mount points are configured on each node as there are streams being used on that node. Each mount point should go to a different disk to ensure that resource contention during media writes is minimal.
-
High quality hardware disks such as SSD or high-speed HDD can lead to improved throughputs, if media-write becomes a bottleneck.
-
We recommend that you configure two nodes configured with the default stream counts in the beginning, and then increase the stream count on the existing nodes or increase the node count itself, if the file server can offer a good read throughput for all the streams.