On Thu, 2018-03-15 at 13:40 -0600, Phillip Chen wrote: > I haven't used the profile argument for FIO before, so when I read > profile I thought it was being used as a synonym I/O profile or > workload. So I think you're right that we shouldn't be creating a new > profile for ZBC but rather focusing on the existing FIO generated > workloads. > When I was talking about valid I meant that get_next_block() would > generate IO that would not cause errors or read filler data past the > write pointer rather than having zbc_adjust_block() modify the IO > afterwards. > You make a good point that the two approaches can easily coexist. If > you get your changes added I would like to try and build the > additional workloads on top of your existing changes so that I can > leverage your write pointer tracking code. Hello Phillip, That sounds like a plan to me :-) Can you have a look at the following code: https://github.com/bvanassche/fio/tree/zbc. The quick fio tests I ran with this code work fine with the ZBC and ZAC disks I tried. I will send that code to Jens once the following two pull requests have been accepted: * "Improvements for analyzing fio with Valgrind" (https://github.com/axboe/fio/pull/560). * "Add an asprintf() implementation" (https://github.com/axboe/fio/pull/561). Thanks, Bart. ��.n��������+%������w��{.n�������^n�r������&��z�ޗ�zf���h���~����������_��+v���)ߣ�
