Travis Downs

sorry forgot the full results: ``` memory DMA alignment: 512 disk DMA alignment: 512 filesystem block size: 4096 context switch per write io (size-changing, append, blocksize 4096, iodepth 1): 0...

> Is this issue fixed completely by ioqueues? Pretty sure the answer is "no". At least in the case where the disk is not performing worse than the io-properties suggest...

Using io_uring may or may not solve this, related question: https://github.com/axboe/liburing/issues/1184

> Do you actually hit nr_request limits? I've seen it with spinning disks, but not SSD/NVMe. Yes, but because the disk (EBS in this case, but it also happens with...

In normal operation, there are only a "few" IOs so we don't hit `nr_requests`. I think this is one of the flaws in the current "feed-forward rate limiting" scheduler (as...

This would be an awesome functionality for writing deterministic programs.

@rockwotj wrote: > As a side note, it looks like `std::string` now (as of C++23) has the ability to be created but uninitialized with [resize_and_overwrite](https://en.cppreference.com/w/cpp/string/basic_string/resize_and_overwrite). It's been around in libc++...

I agree that the 8 bytes increase is probably OK.

> thread_cputime_clock measures both kernel and user time. (Wall time) - (sleep time) overlaps that, since some of the sleep time is spent in the sleeping system call. For epoll/linux-aio...

Here are some results for a really bad case for the existing steal time calculation. This is running `io_tester` with [this patch](https://gist.github.com/travisdowns/740b78b2c59f2282b7e42aa1e88afcab) which prints out steal time metrics periodically and...