Hi, It has been a while since I ran NFS.
You may be able to reduce the ganesha cache with "Entries_HWMark", default it is set to 100000 https://www.mankier.com/8/ganesha-cache-config *Entries_HWMark(uint32, range 1 to UINT32_MAX, default 100000)* The point at which object cache entries will start being reused. https://github.com/nfs-ganesha/nfs-ganesha/issues/377 I have explored to apply with cachinode configuration as below: CacheInode { Attr_Expiration_Time = 600; Entries_HWMark = 50000; LRU_Run_Interval = 90; FD_HWMark_Percent = 60; FD_LWMark_Percent = 20; FD_Limit_Percent = 90; } the above is what I come up after reading the man page. and the result in our test environment, memory usage maintain at ~80%. The work load on the client of this environment is running 4 scripts with the following jobs: * infinitely create 10k&500k files in a loop * infinitely list all created files in a loop * infinitely copy then read a 150k text file * delete all created/copied files every 60sec, The GlusterFS/Ganesha setup and VM specs for the test environment is below: 3 vmware VMs 2 vCPU 4 G memory only 1 volume was shared Before we applied the above settings ganesha.nfsd was killed by oom_killer if the settings when the cacheinode settings above was not loaded after a couple of day that the 4 scripts continuously running. Hope this helps. Regards, Jorick Astrego On 3/21/20 7:29 AM, Strahil Nikolov wrote: > On March 21, 2020 6:34:45 AM GMT+02:00, Olivier <[email protected]> > wrote: >> Hello, >> >> I am running a small Gluster environment with only 4 nodes. And am >> wondering how my gluster machines should be sized: they are 2CPU and >> 4GB >> RAM, but as soon as I connect a client to NFS Ganesha it will start >> swapping like crazy and soon Ganesha will die. >> >> I have been running an NFS server with less than 4GB RAM and 5 or 6 >> clients for years without issue. >> >> Is there a way I can configure both gluster and ganesha to be less >> voracious with RAM? >> >> TIA, >> >> Olivier > Hi Oliver, > > Have you checked if your distribution is building the gluster packages with > the old NFS support? > If it does, then you can use the built-in NFS server which requires less ram. > > About your question, I'm not sure you can control that. Have you tried > using FUSE client on your end systems ? > > Best Regards, > Strahil Nikolov > ________ > > > > Community Meeting Calendar: > > Schedule - > Every 2nd and 4th Tuesday at 14:30 IST / 09:00 UTC > Bridge: https://bluejeans.com/441850968 > > Gluster-users mailing list > [email protected] > https://lists.gluster.org/mailman/listinfo/gluster-users Met vriendelijke groet, With kind regards, Jorick Astrego Netbulae Virtualization Experts ---------------- Tel: 053 20 30 270 [email protected] Staalsteden 4-3A KvK 08198180 Fax: 053 20 30 271 www.netbulae.eu 7547 TA Enschede BTW NL821234584B01 ----------------
________ Community Meeting Calendar: Schedule - Every 2nd and 4th Tuesday at 14:30 IST / 09:00 UTC Bridge: https://bluejeans.com/441850968 Gluster-users mailing list [email protected] https://lists.gluster.org/mailman/listinfo/gluster-users
