3 Prestoserve Administration

This chapter explains how to administer the Prestoserve software. It explains how to select file systems to accelerate and how to use the presto and dxpresto commands to perform the administrative procedures for the day-to-day operation of Prestoserve. It also describes how to check to determine if Prestoserve is working properly.

3.1 Prestoserve Operation

The following sections explain how Prestoserve operates. It describes the Prestoserve buffers and states.

3.1.1 Prestoserve Buffer Management

Prestoserve is implemented as a pseudodevice driver and uses nonvolatile memory to cache synchronous write requests. Write requests are written synchronously to the Prestoserve cache buffers; as the cache fills, old data is written asynchronously to the appropriate disks.

Prestoserve is interposed between other disk drivers and the rest of the Tru64 UNIX kernel. Stubs replace the original driver's entry points in the device switch tables. Whenever Prestoserve needs to perform actual I/O (for example, when the data in the cache needs to be written to disk), it uses the real device driver routines.

Buffers in the Prestoserve cache undergo several phases or states. The buffer transition diagram is roughly as follows:

inval --> dirty --> active --> clean --> dirty

The following list describes the buffer states:

`inval`	An invalid buffer does not presently contain a disk block image.
`dirty`	A dirty buffer contains a valid disk block image that has not yet been written to disk.
`active`	An active buffer is currently in transition to the disk, which means that a write operation has started, but it has not been completed on that buffer.
`clean`	A clean buffer contains a valid disk block image that has been written to disk.

3.1.2 Prestoserve States

The Prestoserve buffer cache is similar to a disk because it contains data. At appropriate times, the data is written to the actual disks. The Prestoserve driver tries to ensure that data is not lost. When a failure occurs, the driver does not discard cache data unless explicitly requested to do so by the system administrator.

Prestoserve is always in one of three states: UP (enabled), DOWN (disabled), or ERROR (error). When the Prestoserve state is UP, Prestoserve improves I/O performance to accelerated file systems by caching synchronous disk write operations to nonvolatile memory. When the Prestoserve state is DOWN, all I/O requests are passed to the actual devices.

Whenever Prestoserve makes a state transition from UP to DOWN, all Prestoserve buffers are successfully flushed (that is, the data is written to disk) and invalidated. If there are dirty buffers in the Prestoserve cache when the system is rebooted, they are flushed, and Prestoserve enters the DOWN state unless an error occurred during the flushing. Some possible disk errors are: the disk drive is write protected or off line, a cable problem exists, or a bad disk block exists.

Note

Because the Prestoserve state is DOWN after a reboot, you may want to set up Prestoserve so that file systems are automatically enabled when the system starts up. Refer to Chapter 2 for information about automatically acclerating file systems.

If an error occurs, Prestoserve enters the ERROR state. When in the ERROR state, the Prestoserve cache is effectively read-only until the error condition is cleared; then, Prestoserve enters the DOWN state. After you fix the disk error, use the presto -u or the presto -U command to verify that the error is corrected. If there are no disk errors, the remaining cached data is written to disk and Prestoserve is reenabled. Refer to Section 3.2.1 for more information about the presto -u and presto -U commands.

The commands that use the reboot() system call cause Prestoserve to enter the DOWN state if all dirty buffers can be successfully flushed. If the buffers cannot be successfully flushed, Prestoserve enters the ERROR state. Commands that are used to reboot the system include the halt, shutdown, and reboot commands. Refer to Chapter 4 for more information on recovering from the ERROR state.

3.2 Managing Prestoserve

The following sections describe how to manage the Prestoserve software. They describe how to select file systems to accelerate, display status, and manage the Prestoserve buffer cache.

The presto command is used to administer Prestoserve. The dxpresto command is used to perform some administrative tasks and also to monitor Prestoserve.

The presto command can perform the following administrative tasks:

Enable and disable file system acceleration

Display information about the accelerated file systems

Display information about Prestoserve state and buffer status

Reset Prestoserve

Write the contents of the Prestoserve cache to disk

Change the size of the Prestoserve cache

Display Prestoserve troubleshooting information

Refer to presto(8), dxpresto(8X), and the following sections for more information.

3.2.1 Accelerating File Systems

Prestoserve can accelerate all mounted file systems on a server, regardless of how many disks or controllers are involved. It is recommended that you accelerate file systems that receive many synchronous write requests. Read-only file systems do not generate synchronous write requests; therefore they are usually not accelerated.

The following list describes some of the types of file systems that may derive benefits from Prestoserve:

File systems that are accessed through the NFS (because many requests for such files are synchronous)

File systems that are used heavily for synchronous I/O

Local file systems (because some operations, such as creating or removing a file, generate synchronous writes)

Remote swapping done to NFS files may benefit from Prestoserve

Prestoserve maintains full block and raw disk semantics. The performance benefits of Prestoserve are not available to raw character device disk partitions. Raw character device reads and writes will flush blocks that are in the Prestoserve cache to disk.

You can use the presto command with the -u or -U option to set the Prestoserve state to UP and enable acceleration on the specified file systems. The -U option sets the Prestoserve state to UP only if the specified directory is the root of a mounted file system. Otherwise, the following message is displayed:

   presto: directory is not a file system root

Note that you can set up Prestoserve to automatically accelerate mounted file systems when the system starts up by specifying the appropriate run-time variables in the /etc/rc.config file and including the file systems in the /etc/prestotab file. Otherwise, you will have to manually accelerate the file systems each time you reboot. Refer to Chapter 2 for more information.

The presto command with the -u or -U option has the following syntax:

presto -u | -U [filesystem]...

Only those file systems specified by the filesystem variable will have Prestoserve enabled. You specify filesystem as a directory mount point (for example, /usr). Do not specify a block device because some functional subsystems, such as the Advanced File System (AdvFS), can map more than one block device to a mount point. If filesystem is not specified, all local writable file systems that are mounted will have Prestoserve enabled. File systems that are presently accelerated will remain accelerated.

If the Prestoserve state was DOWN, the -u and -U options also reset the Prestoserve statistics and buffers to their initial values. If Prestoserve was in the ERROR state, Prestoserve attempts to write to disk any blocks that are still in its cache to make sure that the error has been corrected.

If you mount a local file system using the mount command after the system is running in multiuser mode, you must use the presto -u or presto -U command and specify the mount point to accelerate the file system.

The following examples enable Prestoserve on all mounted read/write local file systems, on a specific mounted file system, and on a directory mount point that is the root of a mounted file system, respectively:

# presto -u 
# presto -u /dsk1g 
# presto -U /usr

3.2.2 Disabling File System Acceleration

You can use the presto command with the -d or -D option to stop Prestoserve acceleration and write any Prestoserve cache data to disk.

The -D option is similar to the -d option, but it sets the Prestoserve state to DOWN only if the specified directory is the root of a mounted file system. Otherwise, the following message is displayed:

    presto: directory is not a file system root

The presto command with the -d or -D option has the following syntax:

presto -d | -D [filesystem]...

Only those file systems specified by the filesystem variable are disabled. You specify filesystem as a directory mount point (for example, /usr). If filesystem is not specified, all accelerated file systems are disabled, and the Prestoserve state is set to DOWN.

Note that the -d and -D options do not reset Prestoserve statistics, and they take effect before the -u, -U, or -R option.

The following command disables the mounted file system /usr:

# presto -d /usr

3.2.3 Displaying the Status of File Systems

You can use the presto command with the -l and -L options to display information about the accelerated file systems.

The -l option lists the accelerated file systems and their mount points in a format that is similar to the mount command.

For example:

# presto -l
/dev/disk/dsk0a on /
/dev/disk/dsk1g on /usr/staff
/dev/disk/dsk2a on /dsk2a
/dev/disk/dsk2g on /dsk2g

The -L option displays any unusual Prestoserve state for the file systems. The unusual states include the following:

`bounceio`	Instead of directly accessing the Prestoserve cache, the disk device receives the data only after it is first copied to main memory.
`disabled`	The file system is not accelerated.
`error`	An error occurred using the file system, and the data has still not been written successfully.

For example:

# presto -L
/dev/disk/dsk0a on /
/dev/disk/dsk0g on /usr (disabled)
/dev/disk/dsk1a on /dsk1a
/dev/disk/dsk1g on /usr/staff
/dev/disk/dsk2a on /dsk2a
/dev/disk/dsk2g on /dsk2g

3.2.4 Displaying the Prestoserve State and Buffer Status

If invoked with no options, the presto command displays the Prestoserve state (either UP, DOWN, or ERROR), the number of bytes of nonvolatile memory the Prestoserve cache is using, the length of time the cache has been enabled, the write cache efficiency, and the current condition of the batteries.

The following is an example of the presto command with no options specified:

# presto 
state = DOWN, size = 0x7e000 bytes
statistics interval: 00:00:00  (0 seconds)
write cache efficiency: 0%
All batteries are ok

You can use the presto command with the -p option to display additional information about the current Prestoserve state and the statistics for write, read, and total operations. The information displayed by the -p option is similar to the information displayed by the dxpresto command.

Example 3-1 shows an example of the presto -p command and its output. A description of the output follows the example.

Example 3-1: Prestoserve Status

# presto -p 
dirty = 0, clean = 61, inval = 0, active = 0
        [1]        [2]          [3]          [4]           [5]           [6]
       count  hit rate  clean hits  dirty hits  allocations  passes
write:  1188     65%       595         182          93        318
 read:     6      0%         0           0           0          6
total:  1194     65%       595         182          93        324
state = UP, size = 0x7e000 bytes
statistics interval: 00:00:35  (35 seconds)
write cache efficiency: 21%      [7]
All batteries are ok         [8]

For each cache read or write operation, Prestoserve increments a counter. A hit occurs when a requested block is matched to a block in a buffer. The previous example shows the following information:

The count specifies the sum of the clean hits, dirty hits, allocations, and passes counters. [Return to example]

The hit rate percentage is the ratio of the clean hits and dirty hits counters to the count.

Note

The hit rate percentage for Prestoserve cache writes indicates the effectiveness of the Prestoserve cache. If the number of read operations is high in proportion to the total count of read and write operations (75% or more), you may improve system performance by increasing the amount of main memory allocated to the file system buffer cache.
[Return to example]

The clean hits counter specifies the number of hits on the clean buffers. [Return to example]

The dirty hits counter specifies the number of hits on the dirty buffers. Each dirty hit represents a physical disk write that was avoided entirely. [Return to example]

The allocations counter specifies the number of new buffers that had to be allocated for disk block images. [Return to example]

The passes counter specifies the number of I/O operations that Prestoserve passed directly to the real device driver. [Return to example]

The write cache efficiency percentage is computed from the ratio of write dirty hits to the number of writes copied into the Prestoserve cache (write count - write passes). [Return to example]

The battery state indicates the condition of the batteries. In general, the battery state can be OK, low, or disabled, but some processors support chargeable batteries and use self-tests to determine if a battery needs charging. If your processor supports chargeable batteries, the battery state can also be specified as in self-test or is charging. [Return to example]

The following is an example of the presto command with the -l and the -p options specified:

# presto -lp 
dirty = 54, clean = 3, inval = 0, active = 4
       count hit rate clean hits dirty hits allocations passes
write:  1236      65%          0        808         421      6
 read:     2       0%          0          0           0      2
total:  1238      65%          0        808         421      8
state = UP, size = 0x7e000 bytes
statistics interval: 00:00:10  (10 seconds)
write cache efficiency: 66%
All batteries are ok
/dev/disk/dsk0a on /
/dev/disk/dsk0g on /usr
/dev/disk/dsk1a on /dsk1a
/dev/disk/dsk1g on /usr/staff
/dev/disk/dsk2c on /dsk2c

3.2.5 Using dxpresto to Administer and Monitor Prestoserve

The dxpresto command starts the worksystem software application that graphically displays information about Prestoserve in a window. You can invoke the dxpresto command on a machine running Prestoserve to obtain that machine's Prestoserve information.

Note

Because dxpresto is a worksystem software application, the DISPLAY environment variable must be set to a machine that is running the worksystem software. See putenv(3) for information on setting environment variables.

The dxpresto command displays the following information:

Prestoserve state

Number of kilobytes of nonvolatile memory that the Prestoserve cache is utilizing

Amount of time that Prestoserve has been enabled

Battery condition

Current state of all the Prestoserve buffers

History of Prestoserve writes per second

History of Prestoserve cache hits per second

Prestoserve statistics for write, read, and total operations

Prestoserve and dxpresto command error messages

The dxpresto command also allows you to modify the displayed information by:

Changing the Prestoserve state to Enabled (UP) or Disabled (DOWN)

Changing the interval of time between Prestoserve queries

Displaying the Prestoserve statistics for write, read, and total operations since Prestoserve was last enabled

Displaying Prestoserve statistics for write, read, and total operations since Prestoserve was last queried

Displaying Prestoserve statistics for write, read, and total operations since a specific time

To invoke dxpresto, use the following command syntax:

/usr/sbin/dxpresto [option]...

You can specify X Window System options with the dxpresto command to control the display and window appearance. See dxpresto(8) for more information.

Note

To use the dxpresto command to monitor a machine's Prestoserve activity, the prestoctl_svc daemon must be running on that machine. Refer to Chapter 2 for information on the prestoctl_svc daemon.

An example of the dxpresto command is as follows:

# dxpresto

Figure 3-1 shows a dxpresto window.

Figure 3-1: dxpresto Window

Figure 3-1 shows the following:

Host

This field displays the name of the host that is running dxpresto.

Presto State

These buttons show the Prestoserve state, either Enabled, Disabled, or Error. If the prestoctl_svc daemon is running, you can toggle the machine's Prestoserve state between Enabled and Disabled. You cannot change an Error state; contact your Compaq Customer Services representative if an Error state occurs.

Sample Interval

This slider shows the interval of time between Prestoserve queries; it allows you to change that interval. When you invoke the dxpresto command, the default Sample Interval is 5; therefore Prestoserve information is gathered every 5 seconds. If you want Prestoserve queried more often, move the slider to the left and click on MB1 until 2 appears, for example; Prestoserve is then queried every two seconds.

Time since last Enable

This field shows the time since Prestoserve was last enabled. The time is displayed in hours, minutes, and seconds and total number of seconds.

Batteries

These graphics show the state of the Prestoserve backup battery system. An intact battery with the word ok indicates that the battery has sufficient power. An intact battery with the word low indicates that the battery's power is low. A broken battery indicates that the battery is disabled. Prestoserve goes into the ERROR state when the backup battery power falls below a minimum amount. Refer to your hardware documentation to determine the minimum amount of backup battery power. Contact your hardware Field Service representative if a battery has insufficient power or is disabled.

Size

This field displays the number of kilobytes of nonvolatile memory that the Prestoserve cache is utilizing. Note that Prestoserve can utilize less than the default maximum size of its Prestoserve cache if you changed the cache size with the presto -s command.

Display Cache Utilization

These buttons allow you to display graphs that demonstrate how the Prestoserve cache is being utilized.

Display Cache Statistics

These buttons allow you to display the cache statistics table.

Exit

This button allows you to exit the dxpresto window.

Message bar

This area at the bottom of the window displays informational and error messages for the dxpresto command and for Prestoserve.

Error messages, such as those indicating RPC communication failure, are displayed on the terminal from which you invoked dxpresto.

Figure 3-2 shows an example of the dxpresto window with both the Display Cache Utilization graphs and the Display Cache Statistics table displayed. The example shows the Writes per second and Hits per second trend line graphs. Each point in the horizontal axis of each graph represents a sample time interval as determined by the Sample Interval slider; the maximum number of samples that can be shown is 210. When you reach the maximum number of samples, the graph shifts to the left so you can see at least the last 105 samples, which is half the maximum number of samples. If you choose 5 in the Sample Interval slider, Prestoserve is queried every 5 seconds; therefore it takes 1050 seconds (5 x 210) to obtain the maximum of 210 samples.

The vertical axis shows the average number of writes performed per second within the sample time interval. For example, if you choose 2 in the Sample Interval slider, Prestoserve is queried every 2 seconds, and each point in the graph shows the average number of writes performed within the interval of 2 seconds. If the graph shows that an average of 5 writes per second were performed within 2 seconds, Prestoserve actually performed 10 writes within those 2 seconds.

Figure 3-2: Expanded dxpresto Window

Figure 3-2 shows the following:

Current Buffer Status: This bar graph shows how the Prestoserve cache operations are distributed among the Prestoserve buffer states, which are described in Section 3.1.1. The vertical axis shows the maximum number of objects or disk blocks that the entire Prestoserve cache can contain. The sum of the four bars is the total number of buffers used in the Prestoserve cache. Note that the size of the Prestoserve cache can be changed by using the presto -s command. See Section 3.3.3 for more information.
Writes per second: This trend line graph shows a recent history of the average number of writes per second over the time intervals that are determined by the Sample Interval slider.
Hits per second: This trend line graph shows a recent history of the average number of Prestoserve cache hits per second over the time intervals that are determined by the Sample Interval slider. The Prestoserve cache hits represent the total number of clean and dirty read and write hits.
Since last Enable: This button allows you to display Prestoserve statistics since Prestoserve was last enabled. This is useful when you want to determine how Prestoserve performs over a long period of time.
Since last Sample: This button allows you to display the Prestoserve statistics for each sample time interval as determined by the Sample Interval slider. If no Prestoserve activity occurs during the time interval, the numbers in the statistics table remain at zero. For example, if the Sample Interval slider is set to 5 and the Since last Sample button is enabled, the statistics table shows the Prestoserve statistics for each interval of 5 seconds.
Since last Zero: This button allows you to display Prestoserve statistics since you clicked on the Zero button. This button allows you to determine how Prestoserve performs over a specific period of time.
Zero: This button sets the numbers in the table to zero, allowing you to specify a time reference for the Prestoserve statistics table. At a later time, you can click on the Since last Zero button to display the Prestoserve statistics since you clicked on the Zero button.
Write Cache Efficiency: This field shows the ratio of write dirty hits to the number of writes copied into the Prestoserve cache.
Prestoserve statistics table: This table is similar to the information that is displayed when you use the presto -p command. For each Prestoserve cache read or write operation, Prestoserve increments an appropriate counter. The table shows the following:
count: Specifies the sum of the clean hits, dirty hits, allocations, and passes
hit rate percentage: Specifies the ratio of clean hits and dirty hits to the total count
clean hits: Specifies the number of hits on the clean buffers
dirty hits: Specifies the number of hits on the dirty buffers (each dirty hit represents a physical disk write that was avoided entirely)
allocations: Specifies the number of new buffers that had to be allocated for the disk block images
passes: Specifies the number of I/O operations that Prestoserve passed directly to the actual device driver

3.3 Handling the Prestoserve Cache

The following sections describe how to write the contents of the cache to disk, how to reset Prestoserve and clear the cache, and how to change the size of the cache.

3.3.1 Writing the Contents of the Cache to Disk

You can use the presto command with the -F option to write the contents of the Prestoserve cache to the available disks but keep the contents of the cache intact.

If the -F option is used and the Prestoserve state is UP, the contents of the cache are written to disk, and the state remains UP. If the Prestoserve state is DOWN, then there is nothing to write to disk, and the state remains DOWN.

If the Prestoserve state is ERROR, as much of the contents of the cache as possible is written to disk. Note that, unlike the -R option, the data in the cache remains after it is written to disk. The state remains ERROR until all the cache data is successfully written to disk. Note that if you cannot write all the cache data to disk and the state remains ERROR, you can use the presto -R command to reset Prestoserve, clear the cache, and set the state to DOWN.

The presto -F command can be used to flush dirty Prestoserve buffers to a disk that was temporarily disabled. For example, if a disk is powered down or disconnected from a bus, the Prestoserve cache could enter the ERROR state. When the disk is again available, you can use the presto -F command to move the cache data to disk and change the Prestoserve state from ERROR to UP.

3.3.2 Resetting Prestoserve and Clearing the Cache

If you are unable to clear the contents of the Prestoserve cache and write the data to disk, you can force Prestoserve out of the ERROR state. You reset Prestoserve and clear the cache by using the presto command with the -R option. The -R option writes as much of the Prestoserve cache data as possible to the appropriate disks, discards the data it cannot write, purges all Prestoserve buffers, and sets the Prestoserve state to DOWN.

Note

The -R option clears the Prestoserve cache by writing the data to the appropriate disks if possible. If a disk is unavailable, the data from the cache is lost, so take care when using this option.

Unlike the -d, -D, and -F options, the -R option discards the Prestoserve cache data that it could not write to disk. The option is useful when cache data is not needed. Note that the -R option takes effect before the -u or -U option.

In the following example, the -R option changes the Prestoserve state to DOWN:


# presto -Rp 
dirty = 0, clean = 61, inval = 0, active = 0
       count hit rate clean hits dirty hits allocations passes
write:  1188      65%        595        182          93    318
 read:    10       0%          0          0           0     10
total:  1198      65%        595        182          93    328
state = DOWN, size = 0x7e000 bytes
statistics interval: 00:00:00  (0 seconds)
write cache efficiency: 0%
All batteries are ok

3.3.3 Changing the Cache Size

You can use the presto command with the -s option to change the size of the Prestoserve cache to the specified number of bytes. The size of the Prestoserve cache is specified in the Prestoserve product description. The presto -s command has the following syntax:

presto -s size

You can specify the size variable using decimal or hexadecimal conventions. For example, both 262144 and 0x40000 represent 256 kilobytes.

To determine how Prestoserve performs with a reduced amount of nonvolatile memory, use the -s option. Note that the size of the Prestoserve cache cannot be larger than the default maximum size or smaller than the default minimum size. If you specify a size that is larger than the default maximum size, the default maximum size is used. If you specify a size that is smaller than the default minimum size, the default minimum size is used. Refer to your processor hardware documentation for information about the default size limits of the Prestoserve cache.

If you specify the -s option and the current Prestoserve state is UP, the state is set to DOWN, the Prestoserve cache is resized, and the state is set to UP.

For example, the following command changes the size of a Prestoserve cache to 512 kilobytes:

# presto -s 0x80000 -p 
dirty = 119, clean = 3, inval = 0, active = 4
       count hit rate clean hits dirty hits allocations passes
write:  1350      66%          0        893         455      2
 read:     0     100%          0          0           0      0
total:  1350      66%          0        893         455      2
state = UP, size = 0x80000 / 0xffc00 bytes
statistics interval: 00:00:00  (0 seconds)
write cache efficiency: 33%
All batteries are ok

3.4 Displaying Debugging Information

You can use the presto command with the -v option to obtain information that you can use to debug Prestoserve operation. The -v option is used with other presto command options and displays extra information to standard output.

3.5 Checking Prestoserve

The system administrator can check to determine if Prestoserve is working properly by performing the following steps:

Log in to the server as root and disable Prestoserve:
```
# presto -d 
```
See Chapter 3 for information on the presto command.

Log in to a client system and mount one of the server's file systems that is exported by the NFS and that has at least as much available space as the size of the client's /vmunix file or some other large file. Use a mount point where the client can create files. The following example uses /usr/tmp as a mount point; the commands establish the client's level of performance without Prestoserve:
```
client% mount  server:/usr/tmp  /mnt
client% cd /mnt
client% /bin/time  cp  /vmunix  bigfile 
  34.1 real         0.0 user         1.1 sys
client% rm  bigfile
```

Enable Prestoserve on the server:
```
server# presto -u
```

Establish the client's level of performance with Prestoserve:

client% /bin/time  cp  /vmunix  bigfile
 10.3 real         0.0 user         1.1 sys
client% rm  bigfile 
client% cd / 
client% umount  /mnt

The real time reported by the commands in step 4 is expected to be about one third of (or about three times faster than) the real time reported by the commands in step 2 while Prestoserve was disabled. Your improvement will vary, but the expected range is between three and five times faster with Prestoserve enabled. If you see much less than a factor of three, make sure that all the other clients are idle and that your network is not being used by others at this time.