I've been waiting for this video ever since the interview Alan did with Wendell of Level1Techs channel, where this presentation was mentioned. I use ZFS on NVME, and the performance really is underwhelming, so anything that could improve the performance is going to pique my interest. Yes, I could switch to mdraid+ext4 and get good performance immediately, but like Alan says at the beginning, once you've gotten used to the ZFS tools anything else feels like going back in time a few decades.
Great presentation! It's really neat hearing what dev work was done during this troubleshooting engagement to create solutions for the client. And now we will all get to benefit from this work eventually. What I love about this talk is hearing the transparency of one of the most vocal ZFS proponents speaking candidly about how it struggled in a scenario and how they are working to overcome it. Great job! And thank you for your candor!
I think it is unwise to think ZFS shouldn't use hardware profiles. There could be detection and customization with an easy to use UX. There aren't that many configs that deserve to be common and selectable. It should be easy and streamline setup across many different hardware with consideration about hardware. It should detect poorly configured SLOGs for example and warn the user.
Flash based SSD's most certainly were a thing in 2001. In fact flash based SSD's predate ZFS by 10 years. It would however be true to say that flash based SSD's were not in common use in 2001.
ouch sound ! at the start a long part without, later very loud cracs (9min15) hopes it gets better ! :)
Sound is really terrible on this, it's detracting from an otherwise pretty good presentation
I agree, the sound of this video is initially pretty bad, but it all cleared up for me at the 11:18 mark.
I've been waiting for this video ever since the interview Alan did with Wendell of Level1Techs channel, where this presentation was mentioned. I use ZFS on NVME, and the performance really is underwhelming, so anything that could improve the performance is going to pique my interest. Yes, I could switch to mdraid+ext4 and get good performance immediately, but like Alan says at the beginning, once you've gotten used to the ZFS tools anything else feels like going back in time a few decades.
Great presentation! It's really neat hearing what dev work was done during this troubleshooting engagement to create solutions for the client. And now we will all get to benefit from this work eventually. What I love about this talk is hearing the transparency of one of the most vocal ZFS proponents speaking candidly about how it struggled in a scenario and how they are working to overcome it. Great job! And thank you for your candor!
Please encourage all ZFS devs to use ZFS root on all their workstations with NVMes! Even if it must be a single disk pool!
Hey that stats lock mentioned at @23:20 was originally reported by my colleague and I a few years back leading to the fix. Fun!
Very good - but in practice which parameters should be changed? It would be important to have a summary and what ranges should be used
I think I now need parity for my ears :(
I think it is unwise to think ZFS shouldn't use hardware profiles. There could be detection and customization with an easy to use UX. There aren't that many configs that deserve to be common and selectable. It should be easy and streamline setup across many different hardware with consideration about hardware. It should detect poorly configured SLOGs for example and warn the user.
Can you share what parameters you change for achieve result? Have NVME's disks with ZFS and have issue with performance.
Flash based SSD's most certainly were a thing in 2001. In fact flash based SSD's predate ZFS by 10 years. It would however be true to say that flash based SSD's were not in common use in 2001.
So like "hardly a thing"
Sounds like we need a PCIe ZFS accelerator card.