4 ms·

For me this is great demonstration of how SSDs have changed the performance landscape. Exactly this idea was discussed on P2D2 10 or so years ago and generally

by dfox 9y ago

For me this is great demonstration of how SSDs have changed the performance landscape. Exactly this idea was discussed on P2D2 10 or so years ago and generally dismissed as too much work for negligible performace gain because the disk IO will still dominate. Today I believe that this makes sense. Edit: and probably it is not only about SSDs, but also about cheap DRAM as today in most deployments I know about most queries only touch indices and tuples that are in RAM (and in fact often the whole pg_base is smaller on disk than RAM of the server).

anarazel 9y ago
I think even 10 years ago that argument was less often true than people might have thought. A raid of a few rotational disks, with multiple controllers, could easily deliver a few hundred MB/s back then too. And thus it was easy to get CPU bottlenecked back then too. The pg community often said so as well, particularly because it was not commonly used in a major analytical capacity...
- fulafel 9y ago
  Also, DRAM hasn't gotten that much cheaper in 10 years. In 2008 you could already get reasonably priced commodity x86 servers with 128 GB RAM, so for most DB applications you could keep all your working set cached and only worry about writes on your storage layer. https://blogs-images.forbes.com/jimhandy/files/2011/12/DRAM-GB-Price.jpg https://blogs-images.forbes.com/jimhandy/files/2011/12/DRAM-... http://thememoryguy.com/wp-content/uploads/2016/05/2016-05-06-Blog-Picture.jpg http://thememoryguy.com/wp-content/uploads/2016/05/2016-05-0... https://en.wikipedia.org/wiki/List_of_Dell_PowerEdge_Servers#Generation_10 https://en.wikipedia.org/wiki/List_of_Dell_PowerEdge_Servers...
- pgaddict 9y ago
  I don't recall the exact discussion from P2D2, but AFAIK the reasoning was more along the lines "There are other bottlenecks that we need to address first." That is, issues that would either limit the JIT gains or issues with better cost/benefit ratio (measured in developer time).

by dfox 9y ago

anarazel 9y ago

I think even 10 years ago that argument was less often true than people might have thought. A raid of a few rotational disks, with multiple controllers, could easily deliver a few hundred MB/s back then too. And thus it was easy to get CPU bottlenecked back then too. The pg community often said so as well, particularly because it was not commonly used in a major analytical capacity...

fulafel 9y ago
Also, DRAM hasn't gotten that much cheaper in 10 years. In 2008 you could already get reasonably priced commodity x86 servers with 128 GB RAM, so for most DB applications you could keep all your working set cached and only worry about writes on your storage layer. https://blogs-images.forbes.com/jimhandy/files/2011/12/DRAM-GB-Price.jpg https://blogs-images.forbes.com/jimhandy/files/2011/12/DRAM-... http://thememoryguy.com/wp-content/uploads/2016/05/2016-05-06-Blog-Picture.jpg http://thememoryguy.com/wp-content/uploads/2016/05/2016-05-0... https://en.wikipedia.org/wiki/List_of_Dell_PowerEdge_Servers#Generation_10 https://en.wikipedia.org/wiki/List_of_Dell_PowerEdge_Servers...
pgaddict 9y ago
I don't recall the exact discussion from P2D2, but AFAIK the reasoning was more along the lines "There are other bottlenecks that we need to address first." That is, issues that would either limit the JIT gains or issues with better cost/benefit ratio (measured in developer time).