How much money for memory?
Paw Prints: Writings of the maddog
Readers may have noticed that I have a theme lately regarding storage sizes, for example talking about disk drives and the TK50 tape drives. This was in anticipation of purchasing a new “laptop” computer, the first new one I have had in six years. More about that machine tomorrow.
Recently I reached into the history museum of my mind and pulled out an old memory board from a VAX 11/780 that we used at Bell Laboratories in the time-frame of 1980. That VAX system used quarter- megabit integrated circuits on the memory boards for its main memory. Since the VAX used ECC correctable memory, it meant that each quarter megabyte of memory had eight chips for the actual data and a chip for the error detection and correction part. To have a whole megabyte of memory on the board you had to have four rows of nine chips each (thirty-six memory chips), plus some chips for the controller. The VAX 11/780 could only have four megabytes of main memory in each system, and Digital's price at the time for memory was about ten thousand dollars a megabyte.
When I first arrived at my Bell Labs location we had only one PDP-11/70 and one VAX 11/780 for the engineers in the building to use. While both machines were rated at about one million instructions per second, the PDP-11 could not hope to compete on address space size or bus speed with the VAX, so I quickly traded it out to a Bell Labs location that needed a PDP-11/70 and got some more VAX machines to take its place.
Each VAX was outfitted with only one megabyte of main memory and one RP06 disk drive (about 180 Mbytes), so I asked to bring each machine up to a full complement of main memory and two more disk drives (to help share the load of accessing the data). These two actions created a forty percent increase in throughput for the systems.
In buying the additional memory we quickly eliminated Digital as a vendor. The two other vendors that we were getting quotes from started at about half the price that Digital was quoting, and we started a bidding war between the two vendors. Finally one vendor gave us a price of one thousand dollars a megabyte and when I told the other salesman the new quote he started crying (real tears) and said that the low price had to reflect a sale where no profit would be made.
Nevertheless we went with the low price. It was also nice that the lower price memory had the chips mounted in sockets, whereas Digital's chips were soldered on the board. Whenever there was one chip that was bad on a board from Digital you had to send the whole board back to be re-worked. When there was one chip bad on the new vendor, all you had to do was replace the bad chip, and you could do that on-site.
Of course you had to tell which chip was bad, and while System V Unix would spit out memory errors that had lots of information, I did not know how to map that information to the chip that was bad on either the board from Digital or the board from our new vendor. Often errors were intermittent, and the board would have to go through substantial testing to find which chip was bad so it could be replaced.
One day I was lamenting the lack of information from hardware vendors (very little has chanced in thirty years) and the salesman for the memory was there with me. He asked why I wanted the information and I told him that if I had the information I could write a program that would tell me exactly which chip was bad on the board so I could replace it myself, or tell him which chip was bad so his service people would not have to guess.
“I will try to get you that information,” he said, “and if you write that program, I will give you a lifetime supply of replacement chips.”
He left, and a few days later I received an email with the information about the board and the chips in it, as well as the way the error messages came out and how it mapped to the chips. I then wrote a program to read in the error message, do the calculations on it, and tell me the board, row and column of which chip was bad. I trained my night operator to run the program, shut the machine down after backups had been taken, open up the back of the system, pull out the correct board, count over the rows and columns to find the chip, and replace it.
When I showed the salesman this happening, he gave me several tubes of quarter-megabit chips and told me that whenever I ran out to just let him know. And of course I gave him a copy of the program.
It was only a couple more years before VAX systems of that type and memory boards of that size were hopelessly obsolete, but Bell Labs saved a considerable amount of money by servicing the memory boards ourselves instead of having to call in the service people of the memory company.
I offered the same thing to my Digital salesman, but he was never able to get the technical specifications for the DEC memory to me.
How does this fit in with my new laptop? It has sixteen gigabytes of main memory in it. If the new laptop was using quarter-megabit memory chips like the VAX did, it would need 576,000 of the chips, and buying the memory from Digital would have cost me 160,000,000 U.S. Dollars.
Carpe Diem!comments powered by Disqus
The company is collaborating with Google and Intel to use Kubernetes as an engine for Fuel
Customers can take a free test drive of SLES for HPC on the Azure Cloud
San Francisco-based chip company announces their first fully open source chip platform.
The whole distro gets rebuilt on glibc 2.3
Ubuntu Vendor tries to solve app packaging and distribution problem across distributions.
Founder of ownCloud launches the Nextcloud project.
Will The Machine change the way future programmers think about memory?
The new Torus distributed storage system is available under an open source license on GitHub
Juries decides Google’s use of Java APIs Was Fair Use