Re: [Qemu-devel] coroutine pool memory usage
i was debugging increased memory footprint of qemu over the past time and found that the coroutine pool heap usage can grow up to 70MB by just booting an Ubuntu Live CD. And those 70MB are never freed. Is this expected? Wouldn't it make sense to asynchronically throw some coroutines (or at least their stack) away if there is no I/O? Yes, perhaps that can be added. But is it RSS that increases, or is it just wasted address space? Paolo
Re: [Qemu-devel] coroutine pool memory usage
Am 27.08.2015 um 17:23 schrieb Paolo Bonzini: i was debugging increased memory footprint of qemu over the past time and found that the coroutine pool heap usage can grow up to 70MB by just booting an Ubuntu Live CD. And those 70MB are never freed. Is this expected? Wouldn't it make sense to asynchronically throw some coroutines (or at least their stack) away if there is no I/O? Yes, perhaps that can be added. But is it RSS that increases, or is it just wasted address space? Sorry, false alarm. I had a debugger running and in this case all memory seems to be allocated. RSS is fine. I tracked the actual stack space that is used to around 1400 Byte per coroutine. So no problem. Peter
[Qemu-devel] coroutine pool memory usage
Hi, i was debugging increased memory footprint of qemu over the past time and found that the coroutine pool heap usage can grow up to 70MB by just booting an Ubuntu Live CD. And those 70MB are never freed. Is this expected? Wouldn't it make sense to asynchronically throw some coroutines (or at least their stack) away if there is no I/O? Does anyone have a pointer to benchmarks of coroutine performance for NAS (iSCSI / NFS) with and without freelist? I would think that it only has significant impact for local (SSD) storage? Thanks, Peter