Skip to main content

Configure clients to distribute ONTAP NAS connections

Contributors netapp-dbagwell
PDFs

To remedy hotspotting, configure the client properly to do its part in preventing CPU bottleneck.

Linux client configuration

Whether you chose an intra-SVM or inter-SVM HDFA deployment, you should use autofs in Linux to make sure clients are load-balancing across the different HDFs. The autofs configuration will differ for inter- and intra-SVM.

Before you begin

You'll need autofs and the appropriate dependencies installed. For help with this, refer to Linux documentation.

About this task

The steps described will use an example /etc/auto_master file with the following entry:

/flexcache auto_hotspot

This configures autofs to look for a file called auto_hotspot in the /etc directory any time a process tries to access the /flexcache directory. The contents of the auto_hotspot file will dictate which NFS server and junction-path to mount inside the /flexcache directory. The examples described are different configurations for the auto_hotspot file.

Intra-SVM HDFA autofs configuration

In the following example, we'll create an autofs map for the diagram in figure 1. Because each cache has the same junction-path, and the hostname cache has four DNS A records, we only need one line:

hotspot cache:/hotspot

This one simple line will cause the NFS client to do a DNS lookup for hostname cache. DNS is setup to return the IPs in a round-robin fashion. This will result in an even distribution of front-end NAS connections. After the client receives the IP, it will mount the junction-path /hotspot at /flexcache/hotspot. It could be connected to SVM1, SVM2, SVM3, or SVM4, but the particular SVM doesn't matter.

Figure 1: 2x2x2 inter-SVM HDFA

Figure 1: 2x2x2 HDFA

Intra-SVM HDFA autofs configuration

In the following example, we'll create an autofs map for the diagram in figure 2. We need to make sure the NFS clients mount the IPs that are a part of the HDF junction-path deployment. In other words, we don't want to mount /hotspot1 with anything other than IP 192.168.0.11. To do this, we can list all four IP/junction-path pairs for one local mount location in the auto_hotspot map.

Note The backslash (\) in the following example continues the entry to the next line, making it easier to read. 
hotspot     cache1:/hostspot1 \
            cache2:/hostspot2 \
            cache3:/hostspot3 \
            cache4:/hostspot4

When the client tries to access /flexcache/hotspot, autofs is going to do a forward-lookup for all four hostnames. Assuming all four IPs are either in the same subnet as the client or in a different subnet, autofs will issue an NFS NULL ping to each IP.

This NULL ping requires the packet to be processed by ONTAP's NFS service, but it doesn't require any disk access. The first ping to return is going to be the IP and junction-path autofs chooses to mount.

Figure 2: 4x1x4 intra-SVM HDFA

Figure 2: s4x1x4 intra-SVM HDFA

Windows client configuration

With Windows clients, you should use an intra-SVM HDFA. To load balance across the different HDFs in the SVM, you must add a unique share name to each HDF. After that, follow the steps in Microsoft documentation to implement multiple DFS targets for the same folder.