On Fri, 2017-06-30 at 18:55 -0400, Greg Dickie wrote: > So I went back and explicitly set compress none. Apparently it > defaults to "client fast". It seems faster now although the size of > the estimate has doubled. What is the deal with compression. > Everything I've seen implies disabling compression on the drive and > using software. Is that the case? >
It supposedly makes estimates more accurate. And cuts down on holding disk and network use. But there's no way you can pull data off one client at 300MB/sec using software compression (maybe if you had a _lot_ of cores and used pigz). It is useful if you're backing up many clients, doing the compression client-side, and you have a holding disk. With your setup, if your data is compressible, you should probably use hardware compression.