Multimedia or Transaction? This only applies to the hardware version of the RAIDION Disk Arrays, also known as Gandiva as mentioned in the following text. This does not apply to RAIDION.plus, RAIDION.es, RAIDION.u2w, etc... RAIDION (Gandiva) Disk Arrays: Transaction code, version 2T43, download 2T43.exe Multimedia code, version 2M43, download 2M43.exe How to upgrade? Read the instructions. Which one to use? Read the technical white paper. Still not sure, then use 2T43. Who Benefits from the Multimedia Codeset? The Multimedia caching scheme is designed to cater to host applications which issue Read/Write requests of sizes 64KB or larger to the Gandiva. The advantage from using these schemes is even greater if the following additional conditions exist: The attendant stripe size on the Gandiva array is large- 64 or 128 blocks. The requests from the host are predominantly sequential. The queue depth for the host requests to the Gandiva is small- preferably 1. Only one host process is accessing the Gandiva for block I/O. When not to recommend the Multimedia codeset: It does not follow that [every] any Multimedia application will benefit from the Gandiva's Multimedia codeset. There are some Multimedia programs that generate Reads and Writes of modest sizes (64 blocks or less), which would find the Gandiva's Transaction caching scheme more suitable for their purposes. Generally, hosts which generate small- and medium-sized file I/O, and frequent file-system updates (e.g., most network servers and database systems) are likely to benefit from the transaction codeset. This is because the request sizes that the Gandiva has to deal with are typically in the 2KB-to-32KB range. Use of the Gandiva's Multimedia codeset of such a scenario will only produce overheads in finding memory resources to process the hosts requests. This would result in the host perceiving lower data- transferrates than expected, and commands taking longer to complete. Usually, multi-user and multi-process systems will find the Gandiva's Transaction Codeset to be a better performer, as they tend to generate random I/O and more frequent commands (especially if the host adapter in the system is making use of the SCSI command-tagging feature). The stripe-size setting on the array, which is also a factor in the overall performance, is usually small for OLTP applications- the default of 8 blocks or 16 blocks. Last Update May 24, 1996