25th Anniversary of the Personal Computer" 4210

one of the things that serial has tended to do is have pair of unidirectional asyncronous connections ... which has helped with various kinds of command and data transfer latencies (more than just the raw bandwidth ... but asyncronous processing allowed more asyncronous overlapping for latency compensation).

both FCS (fiber channel standard) and SCI (scallable coherent interface) provided for asyncronous operation.

circa 1990, 9333s had 80mbit-sec serial copper ... basically doing asyncronous SCSI commands over the serial interfaces ... to potentially a drawer of drives. a single drive was somewhat better than a signle drive scsi ... but four drive configuration had significantly higher aggregate thruput than four drive scsi configuration (in large part because of the asyncronous processing). this eventually evolved into 160mbit SSA (Serial Storage Architecture), i.e. it was so much raw data transfer ... but also minimized latencies by allowing asyncronous operation.

minor reference from long ago and far away

old news article from 1998

SSA technology breaks 3000 IO bottlenecks

from above:

Serial SCSI is the industry's answer to IO problems. In the early 90s, IBM recognized that they not only had to make disks faster, they had to make the interface smarter. They invented and patented two new SCSI interfaces: FibreChannel and SSA. Both use the SCSI command set and are open ANSI standards. Since no vendor yet offers a fibre-channel solution for HP 3000s, this article will discuss the merits of incorporating SSA technology.

... snip ...

part of above may be slightly misstated. 80mbit serial copper was done by Hursley lab (which evolved into SSA).

FCS was different. It was being backed by LLNL ... they had a non-blocking serial copper switch in the mid-80s that they were trying to standardized with 1gbit fiber (fiber channel standard). FCS can also be used for processor-to-processor communication (i.e. tcp-ip can ride on FCS in addition to SCSI commands riding on FCS).

One of the Austin engineers had taken some fiber technology that had been knocking around POK since the 70s ... and enhanced it to run about ten percent faster and use much less expensive optical drivers. The old-time version eventually came out of POK as mainframe "ESCON". The enhanced version came out for the RS-6000 as SLA (serial link adapter). The engineer wanted to then followup with a (proprietary) 800mbit version of SLA. We spent six months convincing him to drop the proprietary effort and instead get involved with the FCS standardization effort (where he became the document editor for the standard).

misc. past posts mentioning FCS, SCI, SSA, SLA, etc.