hckrnws

I once had the privilege of spending a day with Jim Gray, when he visited the National Radio Astronomy Observatory in New Mexico.

At the time, he was interested in collecting as much astronomical data as possible, to create a unified model of the sky.

He viewed it as a logical next step, after creating one of the first unified databases of Earth imagery (Microsoft Terraserver). "I love astronomers," he would sau, "They want to give the data away."

If one can judge a person after a matter of a few hours, then I submit that all of the wonderful things to be said about Jim Gray are true.

.

One story about Terraserver: https://www.vice.com/en/article/8q89q4/microsofts-terraserve...

S.G. Djorgovski1,2 and R. Williams, "Virtual Observatory: From Concept to Implementation", 2005 - PDF: https://authors.library.caltech.edu/28224/1/114.pdf

WorldWide Telescope: https://www.worldwidetelescope.org/

I work in the SQL Server org at Microsoft, though on the service side and not the databasey side. The reverence with which people who knew him talk about Jim Gray is truly awe inspiring.

I don’t have much more to add beyond that. It’s one thing to know someone is a significant force due to their academic legacy, but another to see it in how people who knew them intimately speak of them after they are gone.

Jim Gray wrote a classic paper about fault tolerance that I often reference when talking about Erlang: Why Do Computers Stop and What Can Be Done About It?

http://jimgray.azurewebsites.net/papers/tandemtr85.7_whydoco...

> In the future, hardware will be even more reliable due to better design, increased levels of integration, and reduced numbers of connectors

I couldn't help laughing at that

He did an unofficial follow-up report[0], based on Tandem customer data from 1985-1989. He mentions the big improvements in hardware (at least for Tandem Computers) were the switch to VLSI logic, hard disks that didn't require any maintenance, and the use of fiber optic connections.

I still find Tandem NonStop Systems interesting, and they're still being sold by HPE running on standard x86 servers.

[0] https://jimgray.azurewebsites.net/papers/TandemTR90.1_WhySto...

Better design enabling rowhammer, meltdown, and the like...

But when it comes to failures I would bet things must have improved when you measure failure per operation.

Computers did not fail often 30 years ago. If they failed orders of magnitudes more nowadays we would definitely notice.

I have absolutely no numbers on reliability in any kind of metric.

Gray and Reuter's Transaction Processing: Concepts and Techniques is excellent.

There’s a book reading club in the database internals discord, on this book! We’re up to chapter 8, but a motivated reader can still easily catch up: https://databass.dev/discord

Interesting. I bought the book with Reuter's student discount, but never read it (apart from some pages). Haven't really done anything with databases since university if I don't count looking into some sqlite database an application has stored on my PC or phone.

I guess the book must still be in storage somewhere, I never through away anything...