1.1 Description of Fast Write

Usually when a system goes through a synchronous write for data without using the fast write capability, it goes through the following process:

1. The program issues the write command.

2. The command and data are sent to the SPARCstorage Array controller.

3. The SPARCstorage Array controller sends the command and data to the disk drive.

4. The drive executes the write command.

5. The drive tells the SPARCstorage Array controller that the write was completed successfully.

6. The SPARCstorage Array controller tells the host system that the write was completed successfully.

7. The program issues the next command.

This process can take anywhere from a few milliseconds to tens of milliseconds, because the host system must wait for confirmation from the drive that the write was executed successfully before the program can issue the next command.

If you use the fast write option, however, the data is stored on the NVRAM on the SPARCstorage Array controller for a short period of time. The SPARCstorage Array sends the status back to the system saying that the data was written to the disk (even though it really wasn't), which clears the way for the system to send more information over to the SPARCstorage Array. The SPARCstorage Array continues to store the data on the NVRAM and then flushes all the data from the NVRAM to the disk periodically.

So using the fast write option on a synchronous write for data results in a process similar to the following:

1. The program issues the write command.

2. The command and data are sent to the SPARCstorage Array controller.

3. The SPARCstorage Array controller tells the host system that the write was completed successfully.

4. The program issues the next command.

5. The SPARCstorage Array controller sends the command and data to the disk drive.

6. The drive executes the write command.

7. The drive tells the SPARCstorage Array controller that the write was completed successfully.

In the fast write scenario, if the SPARCstorage Array experiences a power failure after the SPARCstorage Array controller has told the host system that the write was completed (Step 3) and before the drive has told the SPARCstorage Array controller that the write was completed (Step 7), the SPARCstorage Array controller firmware will flush the write to the disk when the array is powered up again.

If the SPARCstorage Array is about to be powered off or if a disk drive is about to be made inaccessible while data is still sitting in the NVRAM, the SPARCstorage Array allows you to flush any outstanding writes from the NVRAM to the drives manually instead of waiting for the array to do it.

In some situations, though, flushing the writes to the drives won't work because the disk drive has failed. If flushing the data to the drive won't work, you will have to purge the data from the NVRAM. The instructions for flushing and purging writes are given later in this manual.

The fast write option is beneficial because it:

• decreases latency for individual commands;
• tends to provide a more consistent response time for write commands;
• can significantly improve performance for applications that do a large amount of synchronous writes, as long as the amount of data being written doesn't overwhelm the capacity of the NVRAM cache.

There are several potential drawbacks to the fast write option, however:

• Management of the fast write NVRAM adds overhead to the SPARCstorage Array controller. So maximum total throughput with fast writes enabled will be lower than it is when fast writes are disabled. (This is true with any write-caching product.)

• If the write to the drive fails for some reason (say, if the drive fails), the application that issued the write will not know that the write failed, as it had been told already that the write succeeded, though an error message is displayed on the system console when this happens. For this reason, it is often recommended that you should use mirroring or RAID-5 so that individual drive failures can be survived without loss of data.

1.2 Description of Reserve/Release

The reserve command allows a host system to reserve individual drives or all the drives in a SPARCstorage Array so that no other host systems can use those drives. This is useful if you have more than one host system connected to a SPARCstorage Array but you want only one system to be able to access certain drives. The release command releases these drives from their reserved state.

1.3 Installing the Software

Refer to the SMCC SPARC Hardware Platform Guide for instructions on installing the software for the SPARCstorage Array. When you have completed the procedures given in that manual, you will have the device drivers, libraries, and the command line interface installed on your system. The device drivers, libraries, and command line interface are required components for the SPARCstorage Array subsystem; you will not be able to operate a SPARCstorage Array without them.

The SPARCstorage Volume Manager is not a required component for the SPARCstorage Array subsystem. However, the Volume Manager gives you many useful disk management and user interface features, including:

• Logical volume management
• Support for disk concatenation, striping, and mirroring
• Online storage reorganization and load balancing
• Visual performance displays
• Graphical user interface (GUI)
• Command line interface (CLI) to the Volume Manager

Because of the disk management capabilities provided by Volume Manager, you may want to use the Volume Manager as a part of the overall SPARCstorage Array subsystem. However, if you do not want to use the

Volume Manager, you can use different disk management software (such as Online: DiskSuite^(TM)), or use the SPARCstorage Array without any kind of disk management software and treat it as a collection of independent SCSI disks.