From: ben@decadent.org.uk 
  
 On Fri, 2025-10-17 at 14:11 +0200, Gero wrote: 
 > Dear experts, 
 > 
 > thanks a lot for your work and commitment on the Debian system. I'm 
 > using Debian for years and I am generally very pleased with it.€€ :-) 
 > 
 > With my company we do numerical simulations and recently did some 
 > benchmarking tests on new AMD EPYC 9334 processors that showed a 
 > significant performance loss of a current Debian system compared to an 
 > older Red Hat or Rocky Linux. We could narrow that down to the following 
 > finding: 
 > 
 > Running these commands: 
 > 
 > cd /dev/shm 
 > echo 3 > /proc/sys/vm/drop_caches 
 > fio --name=random-write --ioengine=posixaio --rw=randwrite --bs=4k 
 > --numjobs=1 --size=4g --iodepth=1 --runtime=60 --time_based --end_fsync=1 
 [...] 
 > I would like to know if you have an explanation or an idea. 
  
 I think for this specific test, the explanation is "this is a stupid I/O 
 pattern that no-one optimises for".  Using AIO with a depth of 1 is 
 effectively doing synchronous I/O in a less efficient way. 
  
 Added to that, POSIX AIO was never that efficient on Linux, and the 
 upstream developers seem to have more-or-less given up on it in favour 
 of io_uring. 
  
 > And I wonder 
 > if you would be interested in investigating the issue any further. Or if 
 > you have a suggestion who I might address preferably. 
  
 If you can also see a regression for io_uring and a more sensible I/O 
 depth then this would probably be interesting for the upstream 
 developers, 
  
 Ben. 
  
 -- 
 Ben Hutchings 
 Computers are not intelligent. They only think they are. 
  
 -----BEGIN PGP SIGNATURE----- 
  
 iQIzBAABCgAdFiEErCspvTSmr92z9o8157/I7JWGEQkFAmj2cJEACgkQ57/I7JWG 
 EQkDQxAAxmggui5KCVKlkQK8zMg+FneICZOZWXqkNNUNFvgWkD0spl1HvqDw2Glj 
 EBqyF/t/MGgHJPhHarkRU2LDvrR7UyJdPtFpggNSXzQIj2XBBcCno1kPCq4VuEwS 
 QHxQ+7Ug3vjl5DLpjtucvWPyBSV9lXRp2T5KPzY6DUVjJDz6Xw4UmDLtHLE0osjt 
 1SrH93f3Nds1WToWXdkVmi7IfpJzHY77PNG0cfZ86OQ62ppHO2AyE+MaXIBGlMjY 
 jN5FbO0UTNtzjP7BDUHIBDh+YEv8kecHPbmMbNPfufxQlrUN9Egx8UUVw46Q11+2 
 Yt9nLBzsAG1vYNwB34B1sYZ9pM6uh/w8fA4yJwrwOE5YVWek9AJKAjyn+OKCZjHx 
 kRaJ9XnLXPaKZA3YANDe37omsVYAe6XsolvzxxhltHFcW0kYJCr3e2SFYDAgWVl6 
 EEDEfXm3nbsujlWbrcdn3DTCPpM3sRZ+uj2QPjRZw3vvmLCcAQMQDMf/TbrVPH3y 
 dQKzxFGm1hCCnCQrXlBAo99sol5rzwulOdDGuI592NE8xksJpzGfenLEHxeguTt4 
 gAcvz/LgYDvPyFdlNkwPJBwpIsX7VFfPgAOyzaWVPSDdJYEHYvPWJET52icmo+ki 
 +rXsXfiUlGTxVWpyur7sax631+oCe9CkXcikVEWF19+DGpMY2H8= 
 =OGYC 
 -----END PGP SIGNATURE----- 
  
 --- SoupGate-Win32 v1.05 
  * Origin: you cannot sedate... all the things you hate (1:229/2)