JDR Computer Products and Electronic Components

Compendium

I/O cards

 

  1. Does the Cache Justify the Cash?
  2. Can a 16 Byte Buffer Make a Difference?
  3. Cache Strategies
  4. RS-232 Standards

<Storage : Table of Contents : Displays>


Does the Cache Justify the Cash?
From Catalog 36, page 29

What is it that we wait on most when we use our computers? The fact is that most computers are I/O bound. We wait the longest when mechanical motion occurs in devices such as floppy, tape and hard drives.

Consider this... If you have a fast hard drive with an average access time of 12 milliseconds and a processor that takes 300 nanoseconds to fetch a byte of memory data, then it takes .012/.0000003 or 40,000 times as long to get that first byte. Subsequent bytes don't do so well either, because we still have to wait for the disk to turn before we can fetch the next byte.

How does a product like the MCT-VCFH or MCT-CIDEFH help? By keeping a copy of frequently requested data in a memory cache, requests for data are frequently answered from the cache and disk activity is minimized. In fact, a 1Mb cache will frequently attain rates of 80% for disk intensive applications.

Back to the Top


Can a 16 Byte Buffer Make a Difference?
From Catalog 29, page 36

You bet it can if the buffer is located in the right place! The 16550 UART has a 16 byte buffer for data that is not only useful, but in many cases, very necessary.

The 16550 is an improved version of the 16450 which itself was an improvement over the 8250 UART. A UART (Universal Asynchronous Receiver Transmitter) is the chip that manages the serial data stream for your COM ports.

When the fastest modems ran at 2400 bps, and PCs only did one thing at a time, the old 8250 and 16450 could buffer long enough for the PC to recognize it and remove it before the next byte arrived to clobber the previous data.

But times change, and so did the modem, the PC, and the way we use them. Today it is common for modems to handle data at 56,000 bps on PCs that are doing several things at once. The old 16450 with its one byte buffer just cannot keep up! Before the PC can switch tasks, recognize the buffer needs emptying, and retrieve that data, several more bytes will arrive and clobber that old small buffer.

Whether you are upgrading or building a new PC, I suggest you select components with 16550 compatibility. Your machine will spend less time correcting errors, and your phone bill will be lower, too.

At the time this was written, I considered the advice important. I now consider it vital. Windows, Windows NT, OS/2 and Unix consume so many processor cycles that responding to an unexpected interrupt fast enough is a huge burden, and frequently will fail to occur in time.

Back to the Top


Cache Strategies
From Catalog 29, page 18

When fighting a war meant hand-to-hand combat, linking up shoulder to shoulder was an appropriate response. However, with the advent of repeating rifles, only a fool would use the same tactic.

With this column, I'm changing my advice on hard disk caches. Not too long ago, I was less than enthusiastic about caching disk controllers. My logic was that memory dedicated to a cache controller couldn't be used for anything else, but main motherboard memory could be used for disk caching when desired, or additional program memory when necessary.

While that is still true, and even appropriate for some users, it isn't nearly as "logical" as it used to be.

Considering the cost of memory, and how much of a good speed improvement you can get from a good sized cache, it can be a good strategy to have dedicated hardware managing your hard drive cache. Improvements of 10:1 are almost assured for any application, and 100:1 or better isn't that rare for reused data.

If you do decide to go for a caching disk controller, don't throw away your old caching program. Additional speed can almost always result if you keep a small (128K to 256K) disk cache in main memory where the CPU can reach it 32 bits at a time.

Back to the Top


RS-232 Standards
From Catalog 12, page 49

Disagree if you must, but I assert that the RS-232 serial communications standard works to the benefit of the computer community. Why then do so many (including me) gripe about its many implementations?

The answer lies in the question. Most applications do not require full implementation in order to work correctly. In fact, it's unusual to find a device, other than a modem, that requires more than four of the nine wires available in the standard PC interface.

Consider the case of a daisy wheel printer with a 2000 byte buffer. What wires do you need to connect to on the PC in order to operate successfully? So that you have the same frame of reference, you need a ground connection. To get the character to type, you need a connection to Transmit Data. To tell the PC you are busy and the buffer is full, you need a signal called a handshake line. Three wires will do it.

What do you do with the other six signals? It turns out you can't completely ignore them. The PC expects a Clear To Send in response to a Request To Send, and Data Terminal Ready in response to a Data Set Ready.

If the handshake line from the printer is connected to a Clear To Send on the PC, then a jumper between Data Set Ready and Data Terminal Ready on the PC's serial connector finishes the job.

The next time you need to connect a serial device, remember what you read here. You don't need to hire an expert to build a special cable, all you need is a little time and patience to get things working. A breakout box with indicator lights will make the job faster.

It has been so long, I almost forget why I wrote this!

Not so many years ago, in a land quite near, PC users trying to save a few dollars would frequently use old teletype machines and serial printers instead of investing in (then) expensive parallel printers.

Because they were from a time before the IBM PC, they usually did not support the standard PC interface and it was not uncommon for the eventual buyer to get the device without schematics. Figuring out the wiring sometimes took considerable time and imagination.

I wonder ... How many of those recycled printers never worked when the buyer got them home?

Back to the Top