Wednesday, September 15, 2021
Parallel is always faster
Ran across a fact I hadn't heard before, in an area that I thought I understood. The fact fits nicely into my latest just-for-fun graphics project on 1957 IBM.
Punched cards are totally old-fashioned. Carrying data around in a box is just hopelessly passe.
Wrong!
Surprising fact: When you need to transfer a REALLY BIG batch of data, it's still easier to do it physically. Big databases or raw unedited video can run into the petabyte range. No matter how good your web connection, it's faster to let FedEx fly the entire set of disks. Amazon has a special product for this purpose, a memory box designed for freight handling.
I hadn't really thought about the question of scale, just ass-u-me-d that big corporations had special web connections. They do, but no web connection can beat a box of memory cards when the box is big enough.
It comes down to the oldest dividing line in communication. Parallel vs serial. A box of cards is parallel. A data wire is serial. Parallel is always faster.
Data communication STARTED as parallel. The Chappe semaphore system was beautifully designed to carry blocks of semantic data as a visual image. It had 256 available positions, most of which were assigned to words and phrases. Later versions even had a code-switch character, denoting that this message will use the Government phrase set or the Stock Exchange phrase set, etc.
Wheatstone's early telegraphs were parallel in the modern sense, with five wires carrying a range of 32 characters.
Morse introduced serial communication, which became the norm for obvious practical reasons. One wire is cheaper to mount and maintain than five wires.
IBM offered web connectors in 1957, using either telegraph** or telephone wires. They transferred about 500 characters per minute.
What was the dividing line in 1957? Let's say you needed to transfer the text of a novel from your office in Chicago to your office in NYC. 100k words = 500k characters, so it would take 1000 minutes or 16 hours. Sending a typed manuscript, or a stack of cards, by truck would take about 20 hours, and a flight would take about 4 hours. So the flight would have been much faster than the web. If you needed to transfer two of those novels or data sets, even the truck would be faster.
Obviously distance is the crucial variable. Between offices in the same city, driving will almost always be faster than any kind of electronic transfer. For a longer distance, the threshold in 1957 was around one megabyte. Now it's around one petabyte, but it's still a threshold.
** "Telegraph" in 1957 was really teletype, using FSK Baudot audio. The direct ancestor of later web modems.
Wrong!
Surprising fact: When you need to transfer a REALLY BIG batch of data, it's still easier to do it physically. Big databases or raw unedited video can run into the petabyte range. No matter how good your web connection, it's faster to let FedEx fly the entire set of disks. Amazon has a special product for this purpose, a memory box designed for freight handling.
I hadn't really thought about the question of scale, just ass-u-me-d that big corporations had special web connections. They do, but no web connection can beat a box of memory cards when the box is big enough.
It comes down to the oldest dividing line in communication. Parallel vs serial. A box of cards is parallel. A data wire is serial. Parallel is always faster.
Data communication STARTED as parallel. The Chappe semaphore system was beautifully designed to carry blocks of semantic data as a visual image. It had 256 available positions, most of which were assigned to words and phrases. Later versions even had a code-switch character, denoting that this message will use the Government phrase set or the Stock Exchange phrase set, etc.
Wheatstone's early telegraphs were parallel in the modern sense, with five wires carrying a range of 32 characters.
Morse introduced serial communication, which became the norm for obvious practical reasons. One wire is cheaper to mount and maintain than five wires.
IBM offered web connectors in 1957, using either telegraph** or telephone wires. They transferred about 500 characters per minute.
What was the dividing line in 1957? Let's say you needed to transfer the text of a novel from your office in Chicago to your office in NYC. 100k words = 500k characters, so it would take 1000 minutes or 16 hours. Sending a typed manuscript, or a stack of cards, by truck would take about 20 hours, and a flight would take about 4 hours. So the flight would have been much faster than the web. If you needed to transfer two of those novels or data sets, even the truck would be faster.
Obviously distance is the crucial variable. Between offices in the same city, driving will almost always be faster than any kind of electronic transfer. For a longer distance, the threshold in 1957 was around one megabyte. Now it's around one petabyte, but it's still a threshold.
** "Telegraph" in 1957 was really teletype, using FSK Baudot audio. The direct ancestor of later web modems.Labels: Morsenet of Things, NOT alternate universe
¶ 12:35 AM
