Twitter | Search | |
Andres Freund (Tech)
Postgres Developer, politics nerd, recently moved to US from Germany. Account about tech related things. For politics:
748
Tweets
75
Following
815
Followers
Tweets
Andres Freund (Tech) Feb 2
The effect can have on my laptop's power usage is just insane. Having some JS heavy website open more than triples power usage.
Reply Retweet Like
Andres Freund (Tech) Jan 31
Don't think so, unfortunately. I hope I'll give an update at another conference later this year - right now this is at a pretty early stage, still...
Reply Retweet Like
Andres Freund (Tech) Jan 31
Reply Retweet Like
Andres Freund (Tech) Jan 31
Slides (somewhat repaired) for my talk Asynchronous IO for PostgreSQL (and probably also Direct IO) are at
Reply Retweet Like
Andres Freund (Tech) Jan 31
Well, that was a bit dramatic. While walking to give the presentation, my laptop shut down. Refused to boot again with something like "No TPM found". Luckily booted after disabling the TPM in bios. Certainly doesn't make for a relaxed start into presenting. Sorry for that!
Reply Retweet Like
Andres Freund (Tech) Jan 27
Replying to @axboe
Update: That's been fixed since. Thanks
Reply Retweet Like
Andres Freund (Tech) Jan 27
Replying to @axboe
I do wonder if it'd be feasible to, optionally, also share sqthreads between urings, not just the wq workers.
Reply Retweet Like
Andres Freund (Tech) Jan 27
Replying to @axboe
I don't think PG can currently benefit from polled submission, due to bottlenecks it has currently, but I hope to address them as part of the io_uring work. And after that using polling e.g. for the journal could be huge.
Reply Retweet Like
Andres Freund (Tech) Jan 27
Replying to @axboe
I'm curious about what those ideas to reduce the synchronization overhead are. Relatedly, one other reason to share urings more widely is that it's feasible to have one or two sqthreads, but not dozens.
Reply Retweet Like
Andres Freund (Tech) Jan 27
Replying to @axboe
The code for that doesn't yet work, but my plan is to have a few urings around that aren't *commonly* going to be shared. So processes not busy doing IO don't hog a full (optimal would be two) urings, the locking is uncontended, but we still can complete their IO if necessary.
Reply Retweet Like
Andres Freund (Tech) Jan 27
Replying to @axboe
Yea, I don't like that overhead either. It's not a bad price to pay to have to do it for journal interleaved writes - the savings are so big. But being restricted in the number of uring instances due to the memlock limits is quite painful for other IO.
Reply Retweet Like
Andres Freund (Tech) Jan 26
Replying to @axboe
Ok, will try to write one. Do you think sharing between processes is a bad path to go down, given the reasoning I outlined ()? I'd rather not work towards building on top of io_uring, that the author of uring finds too constraining for future development.
Reply Retweet Like
Andres Freund (Tech) Jan 26
Replying to @axboe
Cool. Found another fork related one (fcb323cc53e2). I guess I better contribute a liburing test for sharing a uring across a fork... Shoul've upgrade the kernel earlier in the cycle :(
Reply Retweet Like
Andres Freund (Tech) Jan 25
Upgraded Kernel. Now my experimental io_uring branch of postgres is broken :( Different backends, inheriting io_urings from the postmaster, aren't allowed to submit events anymore. Only consume. Hm.
Reply Retweet Like
Andres Freund (Tech) Jan 25
You'll not be there?
Reply Retweet Like
Andres Freund (Tech) Jan 23
Indeed. And it's not just performance in workloads that touch a lot of pages in short order - they can very significantly reduce memory overhead for each connection too.
Reply Retweet Like
Andres Freund (Tech) Jan 16
Replying to @johnregehr
ObjectFile->sections(), SectionRef->isText(), SectionRef->getSize().
Reply Retweet Like
Andres Freund (Tech) Jan 11
Replying to @AndresFreundTec
*but not when benchmarking.
Reply Retweet Like
Andres Freund (Tech) Jan 11
When idle my workstation's power supply regulates down its fan so much it stalls. Which then causes it to go to full speed to "recover". At least I think that's the issue. So to work without undue distraction I now have to trigger load in the background, but not benchmarking.
Reply Retweet Like
Andres Freund (Tech) Jan 7
Replying to @Teggy
postgres' .text is ~8MB.
Reply Retweet Like