The Data Flow analytics tool provides a detailed graphic view of how the IO load is distributed throughout the cluster network over a chosen sampling period within recent days. Using Data Flow, you can monitor the load distribution and identify load balancing issues. Based on the data flow information, you can then take action to adjust VIP pool configurations or client mounts.
For example, you might identify:
VIP balancing issues to be fixed with client remounts.
Rarely used views. Views only appear on the Data Flow page if they received IOs during the selected time period.
Load distribution for top performing users and hosts. For example, if the top n performing users are shown to map to a subset of CNodes.
Overloaded CNodes. Since every client mount is allocated a single VIP and VIPs can move between CNodes, it's possible for a disproportionate number of client users to be writing to the same CNode. It's also possible for clients to mount views using a specific VIP instead of being allocated a VIP by a DNS server. If a single VIP is overused for mounts, that will naturally lead to sub optimal balancing.
Underused CNodes. A CNode that rarely appears on the page is likely to be underused.
Views that receive higher traffic than others. If a specific subset of client hosts is sending most of the traffic to a high-traffic view, you might decide to allocate a VIP pool with a dedicated CNode group to that set of hosts.
Article is closed for comments.