qemu-discuss
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

qemu-user performance issues with big PPC binaries


From: szasz szilard
Subject: qemu-user performance issues with big PPC binaries
Date: Wed, 25 May 2022 09:12:49 +0000 (UTC)

Hello all,
I am trying to run a big realtime multi-threaded ppc (32bit BE) application via QEMU on a x86_64 system (Fedora 35). I have created several test programs that mimic the behavior of the application to test system calls, shared memory and multithreading and to measure emulation performance.
The application and the test programs were compiled with the same gcc options (tried gcc 3.2, gcc4.8.5).
I can only explain the decrease in performance of the main application by its size: a large one, using even several copies of the same .so loaded into memory at several addresses, a large amount of shared memory (IPC), etc.

Has anyone experienced this kind of qemu behavior? Am I referring to the performance differences between small and large binaries. Can this be explained by the advantages of the host processor cache?  I even tried to increase the size of the QEMU blob cache (hashed to 12 bits with) to a ridiculous size (24 bit width) without effect.

Is there a way to explore some statistics from qemu-user ? 

 

Reference PPC machine: 8core VM (ontop via KVM a RH8 ppc64le system-wide OS), RH7.9ppc64,

running on POWER9 host.

Emulator : 4core VM(ontop via KVM a fedora 34 x86_64 system-wide OS), fedora35 x86_64

Running on intel host Intel(R) Xeon(R) CPU E5-2670 0 @ 2.60GHz

 Regards
Szilard.

reply via email to

[Prev in Thread] Current Thread [Next in Thread]