Hello. Thank you very much for your explanation. Because I thought that osd_memory_target will help me limit OSD memory usage which will help prevent memory leak - I tried google and many people talked about memory leak. A nice man, @Anthony D'Atri <aad@dreamsnake.net> , on this forum helped me to understand that it wont help to limit OSD usage. I set it to 1GB because I want to see how this option works. I will read and test with caches options. Nguyen Huu Khoi On Thu, Nov 16, 2023 at 12:23 PM Zakhar Kirpichenko <zakhar@gmail.com> wrote:
Hi,
osd_memory_target is a "target", i.e. an OSD make an effort to consume up to the specified amount of RAM, but won't consume less than required for its operation and caches, which have some minimum values such as for example osd_memory_cache_min, bluestore_cache_size, bluestore_cache_size_hdd, bluestore_cache_size_ssd, etc. The recommended and default OSD memory target is 4 GB.
Your nodes have a sufficient amount of RAM, thus I don't see why you would want to reduce OSD memory consumption below the recommended defaults, especially considering that in-memory caches are important for Ceph operations as they're many times faster than the fastest storage devices. I run my OSDs with osd_memory_target=17179869184 (16 GB) and it helps, especially with slower HDD-backed OSDs.
/Z
On Thu, 16 Nov 2023 at 01:02, Nguyễn Hữu Khôi <nguyenhuukhoinw@gmail.com> wrote:
Hello, I am using a CEPH cluster. After monitoring it, I set:
ceph config set osd osd_memory_target_autotune false
ceph config set osd osd_memory_target 1G
Then restart all OSD services then do test again, I just use fio commands from multi clients and I see that OSD memory consume is over 1GB. Would you like to help me understand this case?
Ceph version: Quincy
OSD: 3 nodes with 11 nvme each and 512GB ram per node.
CPU: 2 socket xeon gold 6138 cpu with 56 cores per socket.
Network: 25Gbps x 2 for public network and 25Gbps x 2 for storage network. MTU is 9000
Thank you very much.
Nguyen Huu Khoi _______________________________________________ ceph-users mailing list -- ceph-users@ceph.io To unsubscribe send an email to ceph-users-leave@ceph.io