The problem, as we’ve seen all to frequently in this industry, is that one part of the technology has progressed faster than another. In this case, the connecting bus has developed at a faster rate than the drives themselves. The result is that, at present, no IDE drive can sustain a transfer rate much greater than 40-45MB/s. That barely even exceeds the bandwidth offered by ATA-33, let alone requiring ATA-100 or ATA-133.
Fig 1. Maxtor's new DiamondMax Plus D740X. The new drive supports the ATA-133 standard.
So if we can’t transfer at even 50MB/s, why bother with ATA-133? It’s not quite that simple. Hard drives have the ability to transmit data in short, fast bursts. Current IDE drives contain a DRAM cache buffer, usually about 2 MB in size, and feature a type of prefetching system. Not totally unlike the prefetching hardware in the Pentium 4 or Athlon XP (although significantly less complicated), the hard drive will fill the cache buffer with data close to the data currently being requested. Frequently, a program requesting that data will also request nearby data, and if it happens to have been stored in the buffer, it can be sent from the cache memory at much greater speeds than it could be read from the platter.
In other words, while the drive may not be able to sustain a transfer rate of 133MB/s, it can certainly burst bits of data at that speed, and probably faster. The question then becomes how often is data burst over the bus? How random versus sequential are today’s programs’ access patterns? This is something we’ll take a look at later.