Chapter 3 NFS Tunable Parameters

This section describes the NFS tunable parameters. For information on kernel tunables, see Chapter 2, Solaris Kernel Tunables. For information on TCP/IP tunables, see Chapter 4, TCP/IP Tunable Parameters.

• NFS Module Parameters

• nfssrv Module Parameters

• rpcmod Module Parameters

Tuning the NFS Environment

You can define these parameters in the /etc/system file, which is read during the boot process. Each parameter can be identified by the name of the kernel module that it is in and a parameter name that identifies it. See Tuning a Solaris System for more information.

---

Note –

The names of the symbols, the modules that they reside in, and the default values can change between releases. Check the documentation for the version of the active SunOS release before making changes or applying values from previous releases.

---

NFS Module Parameters

This section describes parameters relating to the NFS kernel module.

nfs:nfs3_pathconf_disable_cache

Description

Controls the caching of pathconf(2) information for NFS Version 3 mounted file systems.

Data Type

Integer (32–bit)

Default

0 (caching enabled)

Range

0 (caching enabled), 1 (caching disabled)

Units

Boolean values

Dynamic?

Yes

Validation

None

When to Change

The pathconf information is cached on a per file basis. However, if the server can change the information for a specific file dynamically, then use this parameter to disable caching because there is no mechanism for the client to validate its cache entry.

Stability Level

Evolving

nfs:nfs_allow_preepoch_time

Description

Controls whether files with incorrect or negative time stamps should be made visible on the client.

Historically, neither the NFS client nor the NFS server would do any range checking on the file times being returned by using these attributes. The over-the-wire time stamp values are unsigned and 32–bits long, so all values have been legal.

However, on a system running a 32–bit Solaris release, the time stamp values are signed and 32–bits long. Thus, it would be possible to have a time stamp representation that appeared to be prior to January 1, 1970, or pre-epoch.

The problem on a system running a 64–bit Solaris release is slightly different. The time stamp values on the 64–bit Solaris release are signed and 64–bits long. It is impossible to determine whether a time field represents a full 32–bit time or a negative time, that is, one prior to January 1, 1970.

It is impossible to determine whether to sign extend a time value when converting from 32 bits to 64 bits. The time value should be sign extended if the time value is truly a negative number, but should not be sign extended if it does truly represent a full 32–bit time value. This problem is resolved by simply disallowing full 32–bit time values.

Data Type

Integer (32–bit)

Default

0 (32–bit time stamps disabled)

Range

0 (32–bit time stamps disabled), 1 (32–bit time stamps enabled)

Units

Boolean values

Dynamic?

Yes

Validation

None

When to Change

Even during normal operation, it is possible for the time stamp values on some files to be set very far in the future or very far in the past. If access to these files is desired using NFS mounted file systems, then set this parameter to 1 to allow the time stamp values to be passed through unchecked.

Stability Level

Evolving

nfs:nfs_cots_timeo

Description

Controls the default RPC timeout for NFS version 2 mounted file systems using connection oriented transports such as TCP for the transport protocol.

Data Type

Signed integer (32–bit)

Default

600 (60 seconds)

Range

0 to 2³¹ - 1

Units

10th of seconds

Dynamic?

Yes, but the RPC timeout for a file system is set when the file system is mounted. To affect a particular file system, unmount and mount the file system after changing this parameter.

Validation

None

When to Change

TCP does a good job ensuring requests and responses are delivered appropriately. However, if the round-trip times are very large in a particularly slow network, the NFS version 2 client might time out prematurely.

Increase this parameter to prevent the client from timing out incorrectly. The range of values is very large, so increasing this value to be too large might result in real situations where a retransmission was required to not be detected for long periods of time.

Stability Level

Evolving

nfs:nfs3_cots_timeo

Description

Controls the default RPC timeout for NFS version 3 mounted file systems using connection oriented transports such as TCP for the transport protocol.

Data Type

Signed integer (32–bit)

Default

600 (60 seconds)

Range

0 to 2³¹ - 1

Units

10th of seconds

Dynamic?

Yes, but the RPC timeout for a file system is set when the file system is mounted. To affect a particular file system, unmount and mount the file system after changing this parameter.

Validation

None

When to Change

TCP does a good job ensuring requests and responses are delivered appropriately. However, if the round-trip times are very large in a particularly slow network, the NFS version 3 client might time out prematurely. Increase this parameter to prevent the client from timing out incorrectly. The range of values is very large, so increasing this value to be too large might result in real situations where a retransmission was required to not be detected for long periods of time.

Stability Level

Evolving

nfs:nfs_do_symlink_cache

Description

Controls whether the contents of symbolic link files are cached for NFS version 2 mounted file systems.

Data Type

Integer (32–bit)

Default

1 (caching enabled)

Range

0 (caching disabled), 1 (caching enabled)

Units

Boolean values

Dynamic?

Yes

Validation

None

When to Change

If a server changes the contents of a symbolic link file without updating the modification time stamp on the file or if the granularity of the time stamp is too large, then changes to the contents of the symbolic link file might not be visible on the client for extended periods. In this case, use this parameter to disable the caching of symbolic link contents, thus making the changes visible to applications running on the client immediately.

Stability Level

Evolving

nfs:nfs3_do_symlink_cache

Description

Controls whether the contents of symbolic link files are cached for NFS version 3 mounted file systems.

Data Type

Integer (32–bit)

Default

1 (caching enabled)

Range

0 (caching disabled), 1 (caching enabled)

Units

Boolean values

Dynamic?

Yes

Validation

None

When to Change

If a server changes the contents of a symbolic link file without updating the modification time stamp on the file or if the granularity of the time stamp is too large, then changes to the contents of the symbolic link file might not be visible on the client for extended periods. In this case, use this parameter to disable the caching of symbolic link contents, thus making the changes visible to applications running on the client immediately.

Stability Level

Evolving

nfs:nfs_dynamic

Description

Controls whether a feature known as dynamic retransmission is enabled for NFS version 2 mounted file systems using connectionless transports such as UDP. This feature attempts to reduce retransmissions by monitoring server response times, and then adjusting RPC timeouts and read and write transfer sizes.

Data Type

Integer (32–bit)

Default

1 (enabled)

Range

0 (disabled), 1 (enabled)

Dynamic?

Yes, but this parameter is set per file system at mount time. To affect a particular file system, unmount and mount the file system after changing this parameter.

Validation

None

When to Change

In a situation where server response or network load varies rapidly, the dynamic retransmission support might incorrectly increase RPC timeouts or reduce read and write transfer sizes unnecessarily. Disabling this functionality might result in increased throughput, but possibly, also increasing the visibility of the spikes due to server response or network load.

Stability Level

Evolving

nfs:nfs3_dynamic

Description

Controls whether a feature known as dynamic retransmission is enabled for NFS version 3 mounted file systems using connectionless transports such as UDP. This feature attempts to reduce retransmissions by monitoring server response times and then adjusting RPC timeouts and read and write transfer sizes.

Data Type

Integer (32–bit)

Default

0 (disabled)

Range

0 (disabled), 1 (enabled)

Units

Boolean values

Dynamic?

Yes, but this parameter is set per file system at mount time. To affect a particular file system, unmount and mount the file system after changing this parameter.

Validation

None

When to Change

In a situation where server response or network load varies rapidly, the dynamic retransmission support might incorrectly increase RPC timeouts or reduce read and write transfer sizes unnecessarily. Disabling this functionality might result in increased throughput, but possibly, also increasing the visibility of the spikes due to server response or network load.

Stability Level

Evolving

nfs:nfs_lookup_neg_cache

Description

Controls whether a negative name cache is used for NFS version 2 mounted file systems. This negative name cache records filenames that were looked up, but not found. The cache is used to avoid over the network lookup requests made for filenames that are already known to not exist.

Data Type

Integer (32–bit)

Default

1 (enabled)

Range

0 (disabled), 1 (enabled)

Units

Boolean values

Dynamic?

Yes

Validation

None

When to Change

In order for the cache to perform correctly, negative entries must be strictly verified before they are used. This consistency mechanism is relaxed slightly for read-only mounted file systems by assuming that the file system on the server is not changing or is changing very slowly and that it is okay for such changes to propagate slowly to the client. The consistency mechanism becomes the normal attribute cache mechanism in this case.

If file systems are mounted read-only on the client, but are expected to change on the server and these changes need to be seen immediately by the client, then use this parameter to disable the negative cache.

Stability Level

Evolving

nfs:nfs3_lookup_neg_cache

Description

Controls whether a negative name cache is used for NFS version 3 mounted file systems. This negative name cache records filenames that were looked up, but were not found. The cache is used to avoid over-the-network lookup requests made for filenames that are already known to not exist.

Data Type

Integer (32–bit)

Default

1 (enabled)

Range

0 (disabled), 1 (enabled)

Units

Boolean values

Dynamic?

Yes

Validation

None

When to Change

In order for the cache to perform correctly, negative entries must be strictly verified before they are used. This consistency mechanism is relaxed slightly for read-only mounted file systems by assuming that the file system on the server is not changing or is changing very slowly and that it is okay for such changes to propagate slowly to the client. The consistency mechanism becomes the normal attribute cache mechanism in this case.

If file systems are mounted read-only on the client, but are expected to change on the server and these changes need to be seen immediately by the client, then use this parameter to disable the negative cache.

Stability Level

Evolving

nfs:nfs_max_threads

Description

Controls the number of kernel threads that perform asynchronous I/O for the NFS version 2 client. Since NFS is based on RPC and RPC is inherently synchronous, separate execution contexts are required to perform NFS operations that are asynchronous from the calling thread.

The operations which can be executed asynchronously are read for read-ahead, readdir for readdir read-ahead, and write for putpage and pageio requests.

Data Type

Integer (16–bit)

Default

8

Range

0 to 2¹⁵ - 1

Units

Threads

Dynamic?

Yes, but this parameter is set per file system at mount time. To affect a particular file system, unmount and mount the file system after changing this parameter.

Validation

None

When to Change

Change this parameter to increase or reduce the number of simultaneous I/O operations that are outstanding at any given time. For example, for a very low bandwidth network, you might want to decrease this value so that the NFS client does not overload the network. Alternately, if the network is very high bandwidth and the client and server have sufficient resources, you might want to increase this value to more effectively utilize the available network bandwidth and client and server resources.

Stability Level

Unstable

nfs:nfs3_max_threads

Description

Controls the number of kernel threads that perform asynchronous I/O for the NFS version 3 client. Since NFS is based on RPC and RPC is inherently synchronous, separate execution contexts are required to perform NFS operations that are asynchronous from the calling thread.

The operations that can be executed asynchronously are read for read-ahead, readdir for readdir read-ahead, write for putpage and pageio requests, and commit.

Data Type

Integer (16–bit)

Default

8

Range

0 to 2¹⁵ - 1

Units

Threads

Dynamic?

Yes, but this parameter is set per file system at mount time. To affect a particular file system, unmount and mount the file system after changing this parameter.

Validation

None

When to Change

Change this parameter to increase or reduce the number of simultaneous I/O operations that are outstanding at any given time. For example, for a very low bandwidth network, you might want to decrease this value so that the NFS client does not overload the network. Alternately, if the network is very high bandwidth and the client and server have sufficient resources, you might want to increase this value to more effectively utilize the available network bandwidth and the client and server resources.

Stability Level

Unstable

nfs:nfs_nra

Description

Controls the number of read-ahead operations that are queued by the NFS version 2 client when sequential access to a file is discovered. These read-ahead operations increase concurrency and read throughput. Each read-ahead request is generally for 8192 bytes of file data.

Data Type

Integer (32–bit)

Default

4

Range

0 to 2³¹ - 1

Units

Read-ahead requests

Dynamic?

Yes

Validation

None

When to Change

Change this parameter to increase or reduce the number of read-ahead requests that are outstanding for a specific file at any given time. For example, for a very low bandwidth network or on a low memory client, you might want to decrease this value so that the NFS client does not overload the network or the system memory. Alternately, if the network is very high bandwidth and the client and server have sufficient resources, you might want to increase this value to more effectively utilize the available network bandwidth and the client and server resources.

Stability Level

Unstable

nfs:nfs3_nra

Description

Controls the number of read-ahead operations that are queued by the NFS version 3 client when sequential access to a file is discovered. These read-ahead operations increase concurrency and read throughput. Each read-ahead request is generally for 32,768 bytes of file data.

Data Type

Integer (32–bit)

Default

4

Range

0 to 2³¹ - 1

Units

Read-ahead requests

Dynamic?

Yes

Validation

None

When to Change

Change this parameter to increase or reduce the number of read-ahead requests that are outstanding for a specific file at any given time. For example, for a very low bandwidth network or on a low memory client, you might want to decrease this value so that the NFS client does not overload the network or the system memory. Alternately, if the network is very high bandwidth and the client and server have sufficient resources, you might want to increase this value to more effectively utilize the available network bandwidth and the client and server resources.

Stability Level

Unstable

nfs:nrnode

Description

Controls the size of the rnode cache on the NFS client.

The rnode cache, used by both NFS version 2 and 3 clients, is the central data structure that describes a file on the NFS client. It contains the file handle that identifies the file on the server and also contains pointers to various caches used by the NFS client to avoid network calls to the server. Each rnode has a one-to-one association with a vnode. The vnode caches file data.

The NFS client attempts to keep a minimum number of rnodes around to attempt to avoid destroying cached data and metadata. When an rnode is reused or freed, the cached data and metadata must be destroyed.

Data Type

Integer (32–bit)

Default

The default setting of this parameter is 0, which means that the value of nrnode should be set to the value of the ncsize parameter. Actually, any non-positive value of nrnode results in nrnode being set to the value of ncsize.

Range

1 to 2³¹ - 1

Units

rnodes

Dynamic?

No. This value can only be changed by adding or changing the parameter in the /etc/system file, and then rebooting the system.

Validation

The system enforces a maximum value such that the rnode cache can only consume 25% of available memory.

When to Change

Since rnodes are created and destroyed dynamically, the system tends to settle upon a nrnode-size cache, automatically adjusting the size of the cache as memory pressure on the system increases or as more files are simultaneously accessed. However, in certain situations, it might be helpful to set the value of nrnode if the mix of files being accessed can be predicted in advance. For example, if the NFS client is accessing a few very large files, it might be useful to set the value of nrnode to be a small number so that system memory can cache file data instead of rnodes. Alternately, if the client is accessing many small files, it might be helpful to set the value of nrnode large enough to optimize for storing file metadata to reduce the number of network calls for metadata.

Although it is not recommended, the rnode cache can be effectively disabled by setting the value of nrnode to 1. This instructs the client to only cache 1 rnode, which means that it is reused frequently.

Stability Level

Evolving

nfs:nfs_shrinkreaddir

Description

Some older NFS servers might incorrectly handle NFS version 2 READDIR requests for more than 1024 bytes of directory information. This is due to a bug in the server implementation. However, this parameter contains a workaround in the NFS version 2 client.

When this parameter is enabled, the client does not generate a READDIR request for larger than 1024 bytes of directory information. If this parameter is disabled, then the over-the-wire size is set to the minimum of either the size passed in by using the getdents(2) system call or by using NFS_MAXDATA, which is 8192 bytes.

Data Type

Integer (32–bit)

Default

0 (disabled)

Range

0 (disabled), 1 (enabled)

Units

Boolean values

Dynamic?

Yes

Validation

None

When to Change

Examine the value of this parameter if an older NFS version 2 only server is used and interoperability problems are seen when trying to read directories. Enabling this parameter might cause a slight performance drop for applications that read directories.

Stability Level

Evolving

nfs:nfs_write_error_interval

Description

Controls the time duration in between logging ENOSPC and EDQUOT write errors seen by the NFS client. It affects both NFS version 2 and 3 clients.

Data Type

Long integer (32 bits on 32–bit platforms and 64 bits on 64–bit platforms)

Default

5 seconds

Range

0 to 2³¹ - 1 on 32–bit platforms

0 to 2⁶³ - 1 on 64–bit platforms

Units

Seconds

Dynamic?

Yes

Validation

None

When to Change

Increase or decrease the value of this parameter in response to the volume of messages being logged by the client. Typically, you might want to increase the value of this parameter to decrease the number of out of space messages being printed when a full file system on a server is being actively used.

Stability Level

Evolving

nfs:nfs_write_error_to_cons_only

Description

Controls whether NFS write errors are logged to the system console and syslog or to the system console only. It affects messages for both NFS version 2 and 3 clients.

Data Type

Integer (32–bit)

Default

0 (system console and syslog)

Range

0 (system console and syslog), 1 (system console)

Units

Boolean values

Dynamic?

Yes

Validation

None

When to Change

Examine the value of this parameter to avoid filling up the file system containing the messages logged by the syslogd(1M) daemon. When this parameter is enabled, messages are printed on the system console only and are not copied to the syslog messages file.

Stability Level

Evolving

nfs:nfs_disable_rddir_cache

Description

Controls the use of a cache to hold responses from NFS version 2 READDIR and NFS Version 3 READDIR and READDIRPLUS requests. This cache avoids over-the-wire calls to the server to retrieve directory information.

Data Type

Integer (32–bit)

Default

0 (caching enabled)

Range

0 (caching enabled), 1 (caching disabled)

Units

Boolean values

Dynamic?

Yes

Validation

None

When to Change

Examine the value of this parameter if interoperability problems develop due to a server that does not update the modification time on a directory when a file or directory is created in it or removed from it. The symptoms are that new names do not appear in directory listings after they have been added to the directory or that old names do not disappear after they have been removed from the directory.

This parameter controls the caching for both NFS version 2 and 3 mounted file systems. This parameter applies to all NFS mounted file systems, so caching cannot be disabled or enabled on a per file system basis.

Stability Level

Evolving

nfs:nfs3_bsize

Description

Controls the logical block size used by the NFS version 3 client. This block size represents the amount of data that the client attempts to read from or write to the server when it needs to do an I/O.

Data Type

Unsigned integer (32–bit)

Default

32,768 (32 Kbytes)

Range

0 to 2³¹ - 1

Units

Bytes

Dynamic?

Yes, but the block size for a file system is set when the file system is mounted. To affect a particular file system, unmount and mount the file system after changing this parameter.

Validation

None. Setting this parameter too low or too high might cause the system to malfunction. Do not set this parameter to anything less than PAGESIZE for the specific platform. Do not set this parameter too high because it might cause the system to hang waiting for memory allocations to be granted.

When to Change

Examine the value of this parameter when attempting to change the maximum data transfer size. Change this parameter in conjunction with the nfs3_max_transfer_size parameter. If larger transfers are desired, increase both parameters. If smaller transfers are desired, then just reducing this parameter should suffice.

Stability Level

Unstable

nfs:nfs_async_clusters

Description

Controls the mix of asynchronous requests that are generated by the NFS version 2 client. There are four types of asynchronous requests, read-ahead, putpage, pageio, and readdir-ahead. The client attempts to round-robin between these different request types to attempt to be fair and not starve one operation type in favor of another.

However, functionality in some NFS version 2 servers such as write gathering depends upon certain behaviors of existing NFS Version 2 clients. In particular, this functionality depends upon the client sending out multiple WRITE requests at approximately the same time. If one request is taken out of the queue at a time, the client would be defeating this server functionality designed to enhance performance for the client.

Thus, use this parameter to control the number of requests of each type that are sent out before changing types.

Data Type

Unsigned integer (32–bit)

Default

1

Range

0 to 2³¹ - 1

Units

Asynchronous requests

Dynamic?

Yes, but the cluster setting for a file system is set when the file system is mounted. To affect a particular file system, unmount and mount the file system after changing this parameter.

Validation

None. However, setting the value of this parameter to 0 causes all of the queued requests of a particular type to be processed before moving on to the next type. This effectively disables the fairness portion of the algorithm.

When to Change

Change this parameter to increase the number of each type of asynchronous operation that is generated before switching to the next type. This might help with server functionality that depends upon clusters of operations coming from the client.

Stability Level

Unstable

nfs:nfs3_async_clusters

Description

Controls the mix of asynchronous requests that are generated by the NFS version 3 client. There are five types of asynchronous requests, read-ahead, putpage, pageio, readdir-ahead, and commit. The client attempts to round-robin between these different request types to attempt to be fair and not starve one operation type in favor of another.

However, functionality in some NFS version 3 servers such as write gathering depends upon certain behaviors of existing NFS version 3 clients. In particular, this functionality depends upon the client sending out multiple WRITE requests at approximately the same time. If one request is taken out of the queue at a time, the client would be defeating this server functionality designed to enhance performance for the client.

Thus, use this parameter to control the number of requests of each type that are sent out before changing types.

Data Type

Unsigned integer (32–bit)

Default

1

Range

0 to 2³¹ - 1

Units

Asynchronous requests

Dynamic?

Yes, but the cluster setting for a file system is set when the file system is mounted. To affect a particular file system, unmount and mount the file system after changing this parameter.

Validation

None. However, setting the value of this parameter to 0 causes all of the queued requests of a particular type to be processed before moving on to the next type. This effectively disables the fairness portion of the algorithm.

When to Change

Change this parameter to increase the number of each type of asynchronous operation that is generated before switching to the next type. This might help with server functionality that depends upon clusters of operations coming from the client.

Stability Level

Unstable

nfs:nfs_async_timeout

Description

Controls the duration of time that threads, which execute asynchronous I/O requests, sleep with nothing to do before exiting. When there are no more requests to execute, each thread goes to sleep. If no new requests come in before this timer expires, the thread wakes up and exits. If a request does arrive, a thread is woken up to execute requests until there are none again, and then goes back to sleep waiting for another request to arrive, or for the timer to expire.

Data Type

Integer (32–bit)

Default

6000 (1 minute expressed as 60 sec * 100Hz)

Range

0 to 2³¹ - 1

Units

Hz (Typically, the clock runs at 100Hz)

Dynamic?

Yes

Validation

None. However, setting this parameter to a non-positive value has the affect of having these threads exit as soon as there are no requests in the queue for them to process.

When to Change

If the behavior of applications in the system is known precisely and the rate of asynchronous I/O requests can be predicted, it might be possible to tune this parameter to optimize performance slightly in either of the following ways:

• By making the threads expire more quickly, thus freeing up kernel resources more quickly,

• Or, by making them expire more slowly, thus avoiding thread create and destroy overhead.

Stability Level

Evolving

nfs:nacache

Description

Tunes the number of hash queues that access the file access cache on the NFS client. The file access cache stores file access rights that users have with respect to files that they are trying to access. The cache itself is dynamically allocated, but the hash queues used to index into it are statically allocated. The algorithm assumes that there is one access cache entry per active file and four of these access cache entries per hash bucket. Thus, by default, the value of this parameter is set to the value of the nrnode parameter.

Data Type

Integer (32–bit)

Default

The default setting of this parameter is 0, which means that the value of nacache should be set to the value of the nrnode parameter.

Range

1 to 2³¹ - 1

Units

Access cache entries

Dynamic?

No. This value can only be changed by adding or changing the parameter in the /etc/system file, and then rebooting system.

Validation

None. However, setting this parameter to a negative value will probably cause the system to try to allocate a very large set of hash queues, and then hang while trying to do so.

When to Change

Examine the value of this parameter if the basic assumption of one access cache entry per file would be violated. This might be true for systems in a time sharing mode where multiple users are accessing the same file at about the same time. In this case, it might be helpful to increase the expected size of the access cache so that the hashed access to the cache stays efficient.

Stability Level

Evolving

nfs:nfs3_jukebox_delay

Description

Controls the duration of time that the NFS version 3 client waits to transmit a new request after receiving the error, NFS3ERR_JUKEBOX, from a previous request. The error, NFS3ERR_JUKEBOX, is generally returned from the server when the file is temporarily unavailable for some reason. These situations are generally associated with hierarchical storage and CD or tape jukeboxes.

Data Type

Long integer (32 bits on 32–bit platforms and 64 bits on 64–bit platforms)

Default

1000 (10 seconds expressed as 10 sec * 100Hz)

Range

0 to 2³¹ - 1 on 32–bit platforms

0 to 2⁶³ - 1 on 64–bit platforms

Units

Hz (typically the clock runs at 100Hz)

Dynamic?

Yes

Validation

None

When to Change

Examine the value of this parameter and perhaps adjust it to match the behaviors exhibited by the server. The value should be increased if the delays in making the file available are long in order to reduce network overhead due to repeated retransmissions. The value can also be decreased to reduce the delay in discovering that the file has become available.

Stability Level

Evolving

nfs:nfs3_max_transfer_size

Description

Controls the maximum size of the data portion of an NFS version 3 READ, WRITE, READDIR, or READDIRPLUS request. This parameter controls both the maximum size of request that the server returns as well as the maximum size of a request that the client generates.

Data Type

Integer (32–bit)

Default

32, 768 (32 kbytes)

Range

0 to 2³¹ - 1

Units

Bytes

Dynamic?

Yes

Validation

None. Although setting the maximum transfer size on the server to 0 will probably either cause clients to malfunction or just decide not to attempt to talk to the server.

There is also a limit on the maximum transfer size when using NFS over the UDP transport. UDP has a hard limit of 64 kbytes per datagram. This 64 kbytes must include the RPC header as well as other NFS information, in addition to the data portion of the request. Setting the limit too large might result in errors from UDP and communication problems between the client and the server.

When to Change

Change this parameter to tune the size of data being passed over the network. In general, the nfs3_bsize parameter should also be updated to reflect changes in this parameter. For example, when attempting to reduce the default over-the-wire transfer size to 8 kbytes, the value of both the nfs3_max_transfer_size and nfs3_bsize parameters should be changed to 8192 to avoid using multiple operations, each reading or writing 8 kbytes. Alternately, when attempting to increase the transfer size beyond 32 kbytes, then nfs3_bsize should also be updated to reflect the increased value, otherwise no change in the over-the-wire request size is seen.

Stability Level

Unstable

nfssrv Module Parameters

This section describes NFS parameters for the nfssrv module.

nfssrv:nfs_portmon

Description

Controls some security checking that the NFS server can do to attempt to enforce integrity on the part of its clients. It can check to see whether the source port from which a request was sent was a reserved port. This is a port whose number is less than 1024. For BSD based systems, these ports are reserved to processes being run by root. This checking can prevent users from writing their own RPC-based applications to defeat the access checking that the NFS client uses.

Data Type

Integer (32–bit)

Default

0 (checking disabled)

Range

0 (checking disabled), 1 (checking enabled)

Units

Boolean values

Dynamic?

Yes

Validation

None

When to Change

Use this parameter to prevent malicious users from gaining access to files by using the NFS server that they would not ordinarily have access to. However, the reserved port notion is not universally supported. Thus, the security aspects of the check are very weak. Also, not all NFS client implementations bind their transport endpoints to a port number in the reserved range, so interoperability problems might result if the checking is enabled.

Stability Level

Evolving

nfssrv:rfs_write_async

Description

Controls the behavior of the NFS version 2 server when it processes WRITE requests. The NFS version 2 protocol mandates that all modified data and metadata associated with the WRITE request reside on stable storage before the server can respond to the client. NFS version 2 WRITE requests are limited to 8192 bytes of data. Thus, each WRITE request might cause multiple small writes to the storage subsystem. This can cause a performance problem.

One trick to accelerate NFS version 2 WRITEs is to take advantage of a client behavior. Clients tend to send out WRITE requests in batches. The server can take advantage of this behavior by clustering together the different WRITE requests into a single request to the underlying file system. Thus, the data to be written to the storage subsystem can be written in fewer, larger requests. This can increase the throughput for WRITE requests tremendously.

Data Type

Integer (32–bit)

Default

1 (clustering enabled)

Range

0 (clustering disabled), 1 (clustering enabled)

Units

Boolean values

Dynamic?

Yes

Validation

None

When to Change

Some very small NFS clients, particularly PC clients, might not batch WRITE requests. Thus, the behavior required from the clients might not exist, and the clustering in the NFS version 2 server might just add overhead and slow down performance instead of increasing it.

Stability Level

Evolving

nfssrv:nfsauth_ch_cache_max

Description

Controls the size of the cache of client handles that contact the NFS authentication server. This server authenticates NFS clients to determine whether they are allowed access to the file handle that they are trying to use.

Data Type

Integer (32–bit)

Default

16

Range

0 to 2³¹ - 1

Units

Client handles

Dynamic?

Yes

Validation

None

When to Change

This cache is not dynamic, so attempts to allocate a client handle when all are busy will fail. This results in requests being dropped by the NFS server because they could not be authenticated. Most of the time, this is not a problem because the NFS client just times out and retransmits the request. However, for soft-mounted file systems on the client, the client might time out, not retry the request, and then return an error to the application. This might have been avoided by ensuring that the size of the cache on the server is large enough to handle the load.

Stability Level

Unstable

nfssrv:exi_cache_time

Description

Controls the duration of time that entries are held in the NFS authentication cache before being purged due to memory pressure in the system.

Data Type

Long integer (32 bits on 32–bit platforms and 64 bits on 64–bit platforms)

Default

3600 seconds (1 hour)

Range

0 to 2³¹ - 1 on 32–bit platforms

0 to 2⁶³ - 1 on 64–bit platforms

Units

Seconds

Dynamic?

Yes

Validation

None

When to Change

The size of the NFS authentication cache can be adjusted by varying the minimum age of entries that can get purged from the cache. The size of the cache should be controlled so that it is not allowed to grow too large, thus using system resources that are not allowed to be released due to this aging process.

Stability Level

Evolving

rpcmod Module Parameters

This section describes NFS parameters for the rpcmod module.

rpcmod:clnt_max_conns

Description

Controls the number of TCP connections that the NFS client uses when communicating with each NFS server. The kernel RPC is constructed so that it can multiplex RPCs over a single connection, but multiple connections can be used if desired.

Data Type

Integer (32–bit)

Default

1

Range

1 to 2³¹ - 1

Units

Connections

Dynamic?

Yes

Validation

None

When to Change

In general, 1 connection is sufficient to achieve full network bandwidth. However, if TCP cannot utilize the bandwidth offered by the network in a single stream, then multiple connections might increase the throughput between the client and the server.

Increasing the number of connections doesn't come for free though. The price for increasing the number of connections is increased kernel resource usage to keep track of each of the connections.

Stability Level

Evolving

rpcmod:clnt_idle_timeout

Description

Controls the duration of time on the client that a connection between the client and server is allowed to remain idle before being closed.

Data Type

Long integer (32 bits on 32–bit platforms and 64 bits on 64–bit platforms)

Default

300,000 milliseconds (5 minutes)

Range

0 to 2³¹ - 1 on 32–bit platforms

0 to 2⁶³ - 1 on 64–bit platforms

Units

Milliseconds

Dynamic?

Yes

Validation

None

When to Change

Use this parameter to change the time that idle connections are allowed to exist on the client before being closed, if desired. You might might want to close connections at a faster rate to avoid consuming system resources.

Stability Level

Evolving

rpcmod:svc_idle_timeout

Description

Controls the duration of time on the server that a connection between the client and server is allowed to remain idle before being closed.

Data Type

Long integer (32 bits on 32–bit platforms and 64 bits on 64–bit platforms)

Default

360,000 milliseconds (6 minutes)

Range

0 to 2³¹ - 1 on 32–bit platforms

0 to 2⁶³ - 1 on 64–bit platforms

Units

Milliseconds

Dynamic?

Yes

Validation

None

When to Change

Use this parameter to change the time that idle connections are allowed to exist on the server before being closed, if desired. Close connections at a faster rate to avoid consuming system resources, if desired.

Stability Level

Evolving

rpcmod:svc_default_stksize

Description

Sets the size of the kernel stack for kernel RPC service threads.

Data Type

Integer (32–bit)

Default

The default is 0, which means set the stack size to the system default.

Range

0 to 2³¹ - 1

Units

Bytes

Dynamic?

The stack size is set when the thread is created. Therefore, changes to this parameter do not affect existing threads but are applied to all new threads that are allocated.

Validation

None

When to Change

Possibly, very deep call depths can cause the stack to overflow and cause red zone faults. The combination of a fairly deep call depth for the transport, coupled with a deep call depth for the local file system can cause NFS service threads to overflow their stacks.

Set this parameter to a multiple of the hardware pagesize on the platform.

Stability Level

Evolving

rpcmod:svc_default_max_same_xprt

Description

Controls the maximum number of requests that are processed for each transport endpoint before switching transport endpoints. The kernel RPC works by having a pool of service threads and a pool of transport endpoints. Any one of the service threads can process requests from any one of the transport endpoints. For performance, multiple requests on each transport endpoint are consumed before switching to a different transport endpoint. This approach offers performance benefits while avoiding starvation.

Data Type

Integer (32–bit)

Default

8

Range

0 to 2³¹ - 1

Units

Requests

Dynamic?

Yes, but the maximum number of requests to process before switching transport endpoints is set when the transport endpoint is configured into the kernel RPC subsystem. Changes to this parameter only affect new transport endpoints, not existing ones.

Validation

None

When to Change

Tune this number so that services can take advantage of client behaviors such as the clustering that accelerate NFS version 2 WRITE requests. It is possible that increasing this parameter results in the server being better able to take advantage of client behaviors.

Stability Level

Evolving

rpcmod:maxdupreqs

Description

Controls the size of the duplicate request cache that detect RPC level retransmissions on connectionless transports. This cache is indexed by the client network address and the RPC procedure number, program number, version number, and the transaction ID. This cache avoids processing of retransmitted requests that might be non-idempotent.

Data Type

Integer (32–bit)

Default

1024

Range

1 to 2³¹ - 1

Units

Requests

Dynamic?

The cache is dynamically sized, but the hash queues that provide fast access to the cache are statically sized. Making the cache very large might result in long search times to find entries in the cache.

Do not set the value of this parameter to 0. It prevents the NFS server from handling non-idempotent requests.

Validation

None

When to Change

Examine the value of this parameter if false failures are being seen by NFS clients. For example, if an attempt to create a directory fails, but the directory is actually created, it is possible that a retransmitted MKDIR request was not detected by the server.

The size of the cache should match the load on the server. The cache records non-idempotent requests and so only needs to track a portion of the total requests. It does need to hold the information long enough to be able to detect a retransmission on the part of the client. Typically, the client timeout for connectionless transports is relatively short, starting at about 1 second and increasing to about 20 seconds.

Stability Level

Unstable

rpcmod:cotsmaxdupreqs

Description

Controls the size of the duplicate request cache that detects RPC level retransmissions on connection oriented transports. This cache is indexed by the client network address and the RPC procedure number, program number, version number, and the transaction ID. This cache avoids processing of retransmitted requests that might be non-idempotent.

Data Type

Integer (32–bit)

Default

1024

Range

1 to 2³¹ - 1

Units

Requests

Dynamic?

Yes

Validation

The cache is dynamically sized, but the hash queues that provide fast access to the cache are statically sized. Making the cache very large might result in long search times to find entries in the cache.

Do not set the value of this parameter to 0. It prevents the NFS server from handling non-idempotent requests.

When to Change

Examine the value of this parameter if false failures are being seen by NFS clients. For example, if an attempt to create a directory fails, but the directory is actually created, it is possible that a retransmitted MKDIR request was not detected by the server.

The size of the cache should match the load on the server. The cache records non-idempotent requests and so only needs to track a portion of the total requests. It does need to hold the information long enough to be able to detect a retransmission on the part of the client. Typically, the client timeout for connection oriented transports is very long, about 1 minute. Thus, entries need to stay in the cache for fairly long times.

Stability Level

Unstable