Samba - Faster Samba Performance

Samba - Faster Samba Performance

CPU

SMB is single-threaded. This means SMB uses only one CPU core to transfer a file.

See: https://www.samba.org/samba/docs/old/Samba3-Developers-Guide/architecture.html

Usually this is not a problem as SMB does not fully utilize a CPU core. But on extremely busy systems this could overload your CPU core.

A better CPU, especially a multi-threaded one, will make a huge difference.

Bypass single-thread limitation

The single-thread limitation of SMB can be bypassed through opening multiple connections to your server.

Samba Configuration

Samba’s default configuration is fairly poor.

Options which help speed up Samba:

[global]
 
# FORCE THE DISK SYSTEM TO ALLOCATE REAL STORAGE BLOCKS WHEN
# A FILE IS CREATED OR EXTENDED TO BE A GIVEN SIZE.
# THIS IS ONLY A GOOD OPTION FOR FILE SYSTEMS THAT SUPPORT
# UNWRITTEN EXTENTS LIKE XFS, EXT4, BTRFS, OCS2.
# IF YOU USE A FILE SYSTEM THAT DOES NOT SUPPORT UNWRITTEN
# EXTENTS, SET "strict allocate = no".
# NOTE: MAY WASTE DRIVE SPACE EVEN ON SUPPORTED FILE SYSTEMS
# SEE: https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=798532
 
   strict allocate = Yes
 
# THIS IS TO COUNTERACT SPACE WASTAGE THAT CAN BE 
# CAUSED BY THE PREVIOUS OPTION 
# SEE: https://lists.samba.org/archive/samba-technical/2014-July/101304.html
 
   allocation roundup size = 4096
 
# ALLOW READS OF 65535 BYTES IN ONE PACKET.
# THIS TYPICALLY PROVIDES A MAJOR PERFORMANCE BENEFIT.
 
   read raw = Yes
 
# SERVER SIGNING SLOWS THINGS DOWN WHEN ENABLED.
# THIS WAS DISABLED BY DEFAULT PRIOR TO SAMBA 4.
 
   server signing = No
 
# SUPPORT RAW WRITE SMBs WHEN TRANSFERRING DATA FROM CLIENTS.
 
   write raw = Yes
 
# WHEN "strict locking = no", THE SERVER PERFORMS FILE LOCK
# CHECKS ONLY WHEN THE CLIENT EXPLICITLY ASKS FOR THEM.
# WELL-BEHAVED CLIENTS ALWAYS ASK FOR LOCK CHECKS WHEN IT IS
# IMPORTANT, SO IN THE VAST MAJORITY OF CASES,
# "strict locking = auto" OR "strict locking = no" IS ACCEPTABLE.
 
   strict locking = No
 
# TCP_NODELAY:
#    SEND AS MANY PACKETS AS NECESSARY TO KEEP DELAY LOW
# IPTOS_LOWDELAY:
#    [Linux IPv4 Tweak] MINIMIZE DELAYS FOR INTERACTIVE TRAFFIC
# SO_RCVBUF:
#    ENLARGE SYSTEM SOCKET RECEIVE BUFFER
# SO_SNDBUF:
#    ENLARGE SYSTEM SOCKET SEND BUFFER
 
   socket options = TCP_NODELAY IPTOS_LOWDELAY SO_RCVBUF=131072 SO_SNDBUF=131072
 
# SMBWriteX CALLS GREATER THAN "min receivefile size" WILL BE
# PASSED DIRECTLY TO KERNEL recvfile/splice SYSTEM CALL.
# TO ENABLE POSIX LARGE WRITE SUPPORT (SMB/CIFS WRITES UP TO 16MB),
# THIS OPTION MUST BE NONZERO.
# THIS OPTION WILL HAVE NO EFFECT IF SET ON A SMB SIGNED CONNECTION.
# MAX VALUE = 128k
 
   min receivefile size = 16384
 
# USE THE MORE EFFICIENT sendfile() SYSTEM CALL FOR EXCLUSIVELY
# OPLOCKED FILES.
# NOTE: ONLY FOR CLIENTS HIGHER THAN WINDOWS 98/Me
 
   use sendfile = Yes
 
# READ FROM FILE ASYNCHRONOUSLY WHEN SIZE OF REQUEST IS BIGGER
# THAN THIS VALUE.
# NOTE: SAMBA MUST BE BUILT WITH ASYNCHRONOUS I/O SUPPORT
 
   aio read size = 16384
 
# WRITE TO FILE ASYNCHRONOUSLY WHEN SIZE OF REQUEST IS BIGGER
# THAN THIS VALUE
# NOTE: SAMBA MUST BE BUILT WITH ASYNCHRONOUS I/O SUPPORT
 
   aio write size = 16384