On Tue, Dec 02, 2025 at 01:31:40PM +0100, Nicolas George wrote: > [email protected] (HE12025-12-02): > > You are right -- "performance" is a multidimensional thing. I was > > thinking of performance as in "I want to get at many random things > > in those 100T as quickly as possible", your measure was rather > > "I want to get at any of those things as efficiently as possible". > > > > Sometimes you can do things which make those two dimensions happy > > (e.g. better algorithms) and sometimes you have to trade time vs. > > (RAM) space. > > I cannot entirely agree with that. If you throw enough hardware at the > problem to get “at many random things in those 100T as quickly as > possible”, then the system as a whole has good performance because the > hardware has good performance, but it would be wrong to say that the > filesystem has good performance. It is just the hardware.
This is too simplistic, sorry. You need the hardware /and/ a software optimized to take advantage of this hardware. Your software will look wildly different if it "knows" it can count on 16G RAM and a couple of fast cores than if it is designed for ~1G and half-a-laptop core. I won't be tempted to try ZFS on my lowly laptop. I /know/ it'll outperformed by Ext4 any day, hands down (the only credible contender would be XFS, I think). But if I had a 192 core, 512G beast, I'd sure benchmark ZFS on it. Cheers -- t
signature.asc
Description: PGP signature
