4 ms·

Google – A Study In Scalability And A Little Systems Horse Sense

62 points by abraham 16y ago

btilly 16y ago
For those who don't know, Jeff Dean is probably the most respected developer within Google. See http://research.google.com/people/jeff/index.html http://research.google.com/people/jeff/index.html for some of what he has done.
- moultano 16y ago
  Internally there's a "Jeff Dean Facts" page in the spirit of Chuck Norris Jokes. It includes things such as "Once, in early 2002, when the index servers went down, Jeff Dean answered user queries manually for two hours. Evals showed a quality improvement of 5 points."
  - mdwrigh2 16y ago
    Is there any chance of this being released to the public? ...or even a chance of you sending me a copy (which of course I would keep to myself!)
    - moultano 16y ago
      It references too many internal things for that, but as a consolation prize, here's my second favorite: "Jeff Dean puts his pants on one leg at a time, but if he had more than two legs, you would see that his approach is actually O(log n)."
elblanco 16y ago
Absolutely fantastic notes, the standout ones to me: "They have had 7 significant revisions in 11 years" - This echos something I've learned in my own career. You'll never ever ever ever get it right the first time. If you spend all of your time trying to get it right the first time, you'll never produce anything. BUILD! (but plan well to eliminate obvious mistakes). "All their systems work well inside a datacenter, but have no way f spanning datacenters." - This was a real surprise. I guess I had mistakenly assumed that all of Google's distributed resources lived as one giant meta-machine. I never realized how disconnected the datacenters were. "Don’t design to scale infinitely – consider 5X – 50X growth. But > 100X requires redesign << very insightful" - quote says it all.
- btilly 16y ago
  If you care about latency, you can't abstract away the existence of data centers. The round trip speed of light means that that abstraction leaks too much.
  - nostrademons 16y ago
    Such a system could, in theory, abstract away the existence of data centers by optimizing for latency. Wherever the user is, perform the computation by using the copies of data that result in the lowest time spent hopping between DCs. And shard the data between DCs so that data that's frequently used together stays together, and data lives near its likely point of use, and replicas give good coverage of possible access points. Performing this calculation in a way that's faster than just going and fetching the data is left as an exercise to the reader. And Jeff Dean.

62 points by abraham 16y ago

For those who don't know, Jeff Dean is probably the most respected developer within Google. See http://research.google.com/people/jeff/index.html http://research.google.com/people/jeff/index.html for some of what he has done.

moultano 16y ago
Internally there's a "Jeff Dean Facts" page in the spirit of Chuck Norris Jokes. It includes things such as "Once, in early 2002, when the index servers went down, Jeff Dean answered user queries manually for two hours. Evals showed a quality improvement of 5 points."
- mdwrigh2 16y ago
  Is there any chance of this being released to the public? ...or even a chance of you sending me a copy (which of course I would keep to myself!)
  - moultano 16y ago
    It references too many internal things for that, but as a consolation prize, here's my second favorite: "Jeff Dean puts his pants on one leg at a time, but if he had more than two legs, you would see that his approach is actually O(log n)."

elblanco 16y ago

Absolutely fantastic notes, the standout ones to me: "They have had 7 significant revisions in 11 years" - This echos something I've learned in my own career. You'll never ever ever ever get it right the first time. If you spend all of your time trying to get it right the first time, you'll never produce anything. BUILD! (but plan well to eliminate obvious mistakes). "All their systems work well inside a datacenter, but have no way f spanning datacenters." - This was a real surprise. I guess I had mistakenly assumed that all of Google's distributed resources lived as one giant meta-machine. I never realized how disconnected the datacenters were. "Don’t design to scale infinitely – consider 5X – 50X growth. But > 100X requires redesign << very insightful" - quote says it all.

btilly 16y ago
If you care about latency, you can't abstract away the existence of data centers. The round trip speed of light means that that abstraction leaks too much.
- nostrademons 16y ago
  Such a system could, in theory, abstract away the existence of data centers by optimizing for latency. Wherever the user is, perform the computation by using the copies of data that result in the lowest time spent hopping between DCs. And shard the data between DCs so that data that's frequently used together stays together, and data lives near its likely point of use, and replicas give good coverage of possible access points. Performing this calculation in a way that's faster than just going and fetching the data is left as an exercise to the reader. And Jeff Dean.