Geoffrey MEZERETTE
2019-Feb-05 13:20 UTC
[Samba] Question about Sizing of Samba Linux Server
Hello all, I'm jumping into this mailing list because my question still unsolved after Googling the web. Do you know any performance of sizing recommendations about Samba 4.7.1 ? For few informations : 2vCPUs / 1 Core 8Go of RAM 140 simultaneous Windows users / 250 at most 20 NFS shares mounted via fstab This question because my RAM is all eated up in the buffer/cache (normal according to last chapter of : https://wiki.samba.org/index.php/Performance_Tuning). Users are not complaining but I don't know if the servers is needing more or less. Any advice will be greatly appreciated, thanks. *Geoffrey MEZERETTE* • Ingénieur Systèmes et Réseaux • Enovea • 16, rue Jacques Monod • 76130 Mont Saint Aignan • Standard +33 (0) 2 32 18 82 66 • geoffrey.mezerette at enovea.net • www.enovea.net <http://www.enovea.net/>• <http://www.enovea.net/>
J Martin Rushton
2019-Feb-05 14:40 UTC
[Samba] Question about Sizing of Samba Linux Server
Memory taken up by the buffer cache is available for use. On a Linux system writes to disk occur asynchronously, th e process first moves the data into system buffers, then continues with what it was doing. At a later time the operating system moves the data from memory to disk, but does not immediately blank the page. Pages which are left as a copy of disk are therefore a cache, but can be released if another process requires them. Indeed, they are _only_ released when required, so that over time all memory pages end up being used, or dwelling in the buffer cache. Cached pages are recovered very quickly with a soft page fault, going to disk requires a hard page fault which is dealt with a disk speeds. Be aware that the foregoing only applies to single node machines (including those servering the network via NFS of Samba). In a true cluster environment with a SAN, the on-disk version may be updated by another node, so the cluster-wide filesystem needs to invalidate the cache, but that is a whole new level of complexity. The short answer therefore is that "my RAM is all eated up in the buffer/cache" is indeed normal. In general, more RAM is nearly always better, but if "users are not complaining" you may have difficulty getting past management! HTH, Martin On 05/02/2019 13:20, Geoffrey MEZERETTE via samba wrote:> Hello all, > > I'm jumping into this mailing list because my question still unsolved after > Googling the web. > > Do you know any performance of sizing recommendations about Samba 4.7.1 ? > > For few informations : > > 2vCPUs / 1 Core > 8Go of RAM > 140 simultaneous Windows users / 250 at most > 20 NFS shares mounted via fstab > > > This question because my RAM is all eated up in the buffer/cache (normal > according to last chapter of : > https://wiki.samba.org/index.php/Performance_Tuning). > > Users are not complaining but I don't know if the servers is needing more > or less. > > Any advice will be greatly appreciated, thanks. > > *Geoffrey MEZERETTE* • Ingénieur Systèmes et Réseaux • Enovea • > 16, rue Jacques Monod • 76130 Mont Saint Aignan • Standard +33 (0) 2 32 18 > 82 66 • geoffrey.mezerette at enovea.net • > www.enovea.net <http://www.enovea.net/>• <http://www.enovea.net/>
Seemingly Similar Threads
- Troubles converting a pv host from dom0-hosted kernel to self-contained kernel
- nonlinear curve fit of an implicit function
- cmd-vacation.c:4:17: fatal error: lib.h: No such file or directory && ./configure: line 11410: -lssl: command not found
- Asterisk processes
- Hdd maximum size