Improving filer performance

About this section

This section describes configuration procedures that might improve your filer's performance.

Balancing NFS traffic on network interfaces

Attach multiple interfaces on the filer to the same physical network to balance network traffic among different interfaces. For example, if two Ethernet interfaces on the filer named toaster are attached to the same network where four NFS clients reside, specify in /etc/fstab on client1 and client2 that these clients mount from toaster-0:/home. Specify in /etc/fstab on client3 and client4 that these clients mount from toaster-1:/home. This scheme can balance the traffic among interfaces if all clients generate about the same amount of traffic.

The filer always responds to an NFS request by sending its reply on the interface on which the request was received.

Ensuring reliable NFS traffic by using TCP

With faster NICs and switches, NetApp recommends supporting NFSv2 or NFS v3 protocol over TCP rather than over UDP. NFS v4 is supported over TCP only.

Avoiding access time update for inodes

If your applications do not depend on having the correct access time for files, you can disable the update of access time (atime) on an inode when a file is read. Set the no_atime_update option to On for each volume to prevent updates. For example:

vol options vol2 no_atime_update on

Consider turning this option on if your filer has extremely high read traffic--for example, on a news server used by an Internet provider--because it prevents inode updates from contending with reads from other files.

Caution
If you are not sure whether your filer should maintain an accurate access time on inodes, leave this option set at its default, Off, so that the access time is updated.

Improving read-ahead performance

If the file access patterns of your clients are random (nonsequential) and the cache age is less than 3, setting minimal read-ahead to On might improve performance. By default, the filer uses aggressive read-ahead, which enhances sequential access, and is more commonly used by UNIX clients and applications. Use the sysstat command to determine the cache age. Set the minra option to On for each volume to specify minimal read-ahead. By default, the option is set to Off and the filer does very aggressive read-ahead.

Adding disks to a disk-bound volume

If you have a single-volume filer, use the sysstat -u command to determine the fraction of time that the busiest disk is active. If the fraction is greater than 80%, add disks to the volume through the vol add command.

Making optimal use of Slot 8

This method of improving filer performance applies to F700-series filers only. The fastest-available PCI slot in an F700-series filer that is not clustered is Slot 8. If your filer uses a 64-bit network interface card, put it into Slot 8 to take advantage of the speed. If the filer does not use a 64-bit network interface card, put the main FC-AL adapter into this slot.

Putting cards into the correct slots

At boot time or when you use the sysconfig -c command, you might see messages indicating that expansion cards must be in certain slots. To improve performance, follow the recommendations in the message. See the System Configuration Guide at http://now.netapp.com/NOW/knowledge/docs/hardware/syscfg/ for recommended card placement.

Maintaining adequate free blocks and free inodes

If the percentage of free blocks or free inodes falls to less than 10% on any volume, the performance of writes and creates can suffer. Check free blocks and inodes by using the df command and df -i command, respectively. If the percentage of used blocks is greater than 90%, increase blocks by adding disks or deleting snapshots. If the percentage of free inodes is less than 10%, increase inodes by deleting files or using the maxfiles command.

Using oplocks for CIFS filers

Oplocks allow CIFS clients to cache more data locally, reducing traffic to the filer. On a filer running CIFS that is not in a database environment, set oplocks to On by using the options cifs.oplocks.enable on command. If the filer is in a database environment, ensure that the oplocks are not set.

Increasing the TCP window size for CIFS

Increasing the TCP receive window size to its maximum setting on both the filer and a CIFS client can improve performance for large transfers, provided that packet loss is not taking place and the client's send window is large.

Caution
Network Appliance recommends that you call Network Appliance Technical Support before changing this value.

The default is 17,520 bytes. The number of bytes must be between 2,920 and 64,240, in multiples of 1,460 bytes. The TCP window size controls the number of TCP messages that can be transmitted between the filer and the CIFS client at one time.

Caution
If you are using DOS, enter an NT4 value of 8,760.

Use the options cifs.tcp_window_size 64240 command to maximize the TCP window size on a filer running CIFS. Use the options nfs.tcp.recvwindowsize 64240 command to maximize the TCP window size on a filer running NFS. Change the window size in the Windows registry on a Windows NT client by adding the DWORD value \\HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Services\ Tcpip\Parameters\TcpWindowSize and set it to 64,240 (0xFAF0 in hexadecimal).

Note
The cifs.tcp_window_size and nfs.tcp.recvwindowsize options are invisible until you set them. After you set these invisible options, you can view them by entering the options cifs or the options nfs command.

Backing up by using qtrees

If your filer has multiple tape drives and a volume that has two to four qtrees, you can improve backup rates by running multiple dump commands in parallel, each reading a different qtree and writing to a different tape drive.