Serial ATA, more than a fashion statement

ASK WHO YOU might and probably anybody who ever stuck his or her nose into a PC will tell you the same story: Ribbon cables are ugly and clumsy. Forty wires to connect a storage device to a host controller and only 16 are really needed for data transfer of two Bytes at a time. The remaining 28 wires are only there for support, meaning that there are commands and there are status reports along with some grounding. Because this was still not enough, we just recently had to undergo a doubling of the wires to a total of 80 but that was only to add extra shielding.

When the original cables were designed, they were built to transfer 3 MB/sec and now we are running at 133 MB/sec, 44 times faster than what the inventors thought would be reasonable and still using the same technology. Needless to say that there had to be some evolution, even in what appears a simple cable.

Darwinism has shown mere growth often leading to extinction, extra brain power on the other hand has always been an asset. The same goes for storage media where the probably most significant improvement was the addition of a small amount of memory to the drive itself to make data transfer independent from the actual read/write process from/to the media. Consequently, data could be prefetched, stored inside this "cache" and delivered upon request in one burst at the maximum speed supported by the interface.

Caching the data would still be a futile effort if the transfers needed to pass through the CPU in order to be written to memory. The solution was a new technology, allowing direct data transfer from the HDD to the system memory, bypassing any intermediate processing by the CPU. This technology is called "direct memory access" or DMA. The result was an incredible 33 MB/sec UltraDMA transfer.

While controllers were getting faster and the memory was getting faster, too, the original ribbon cables could no longer maintain the signal integrity. Electrical cross-talk between neighboring data lines where each pair of line acts like a transformer became Enemy #1 by generating false data inside the cable. The laws of physics could be tricked, once again by inserting the above mentioned shield wires and lowering the signal voltage swing from 5V to 3.3V.

The laws of physics also state that no two bodies can occupy the same space and if the number of cables is doubled, their diameter has to decrease accordingly. Smaller calibers, in turn make for slower transmission and chances are that there will be a mismatch somewhere causing a propagation delay between individual bits of the same transfer. Increasing the drive strength of the signal will iron out some of the bumps but at the same time there is a chance for naughty ground bouncing, signal ringing and some electrical equivalent of a sonic boom in the cable. Neither of its ends will like that.

What clumsiness never achieved was finally done by the limitations in signaling, Parallel ATA is dying. The new design is sleek and flexible and listens to the name of Serial ATA and mind me saying that the rainbow spectrum of the cables is just one aspect. Serial signaling only requires a single signal line, full duplex will need two and differential signaling will need two wires each for input and output. Low Voltage Differential Signaling (LVDS) using full duplex I/O will, therefore, only need four data wires but it needs some control data integrated into the signal stream to avoid signal disparity, hence we got ten bits to encode one Byte.

Jurassic Park taught us that Dinos were not just passive but Jurassic Park came out after Parallel ATA was invented. Not that it matters since PATA drives are just what they are and that is passive, meaning that they can only answer when asked but not actively communicate data. A read begins with a request from the host bus adapter upon which the drive starts to collect the data needed. Once the drive is ready it'll flip the SERVICE bit. The next time the host polls the drive for any status changes and sees that the drive is, indeed, ready it will issue a SERVICE command and the drive can start transmitting data.

Serial ATA drives, on the other hand, are permanently connected to the host using point to point mapping and as soon as the drive has the data or only a fraction thereof, it can start the transfer using what is called "FirstPartyDMAsetup" and "out of order data delivery". The benefit of this mechanism of drive-initiated transfer is threefold: First, there are no idle periods. Second, while retrieval of the data is limited by the mechanical properties of the drive, piecing the individual fragments from out of order data transfers back together occurs at the speed of the system processor and is, therefore, orders of magnitude faster. Third, since only ½ of the commands have to be issued by the processor, the CPU utilization is generally lower.

In some benchmarks we have already seen up to a doubling of the performance within the same drive just by switching from a parallel to a serial interface, most likely enabled by the reduced command overhead but this is only the beginning. The current SATA1.5GB specs will extend to SATA3.0 and SATA6.0 within the next three years maintaining full backward compatibility. Additional features like native command queuing will allow the drive itself to make intelligent decisions regarding the order of executing outstanding commands. This will further speed up the drives and at the same time reduce wear and tear because of optimization of the workload. All in all, SATA is here to stay.

Michael Schuette is editor-in-chief of the very excellent Lost Circuits