On May 14, 2020, at 09:26, Joerg Sonnenberger <[email protected]> wrote: > > On Wed, May 13, 2020 at 10:11:14PM -0400, John Franklin wrote: >> There are scalability issues with Mercurial, too. I cloned NetBSD src >> on a 1GB RAM, 1GB swap, 4 CPU VM (Debian Buster) using git from the >> GitHub project and from anonhg.netbsd.org. > > You are comparing Apples and Oranges. The clonebundles on anonhg are > created with very large windows and zstd compression and the necessary > buffering of that is the primary memory use. I used to provide bzip2 > bundles as fallback, but disk space constrains made that temporarily > undesirable. I.e. this is not about scaling at all.
I’m comparing cloning with cloning using the same VM. From the contributor’s POV, it's as apples-to-apples as it gets. Even the commands are the same: “$VCS clone $URL” As configured, Mercurial takes 3x the memory, 4x the time, and fails to clone at all without a minimum of 3GB of RAM+swap. The driving factor behind the resource requirement is the amount of history the project has. Which is to say, how well these two tools *scale* with the size of the repository. If server-side changes can reduce that, and all the server needs is a bigger disk, then get a bigger disk. jf -- John Franklin [email protected]
