RPC HI/LO Marks

How can I increase the default RPC_HI_MARK?
What value should the default send buffer be set to, for best Performance?

With the adaptive RPC_HI_MARK feature added to the 2003.1 server (change 43934), best performance is obtained (especially in proxy configurations) when the default send buffer size on the server machine is 32K or greater. In most modern operating systems, the default send buffer size is 32K. The notable exception is in Linux 2.4 kernels, where the default send buffer size is 16K.

The default send buffer size on Linux 2.4 kernels can be increased from 16K to 32K either dynamically or at boot-time. For example, the following dynamically changes the default send buffer size on the Perforce Server machine:

# /sbin/sysctl net.ipv4.tcp_wmemnet.ipv4.tcp_wmem = 4096 16384 131072# /sbin/sysctl -w net.ipv4.tcp_wmem="4096 32768 131072"net.ipv4.tcp_wmem = 4096 32768 131072# /sbin/sysctl net.ipv4.tcp_wmemnet.ipv4.tcp_wmem = 4096 32768 131072

The following example changes default send buffer size at boot-time on the Perforce Server machine:

# /sbin/sysctl net.ipv4.tcp_wmemnet.ipv4.tcp_wmem = 4096 16384 131072# vi /etc/sysctl.conf...# p4 diff /etc/sysctl.conf== //depot/machines/linux/etc/sysctl.conf#1 - /etc/sysctl.conf ==17a18,20>> # Increase default send buffer size for Perforce performance> net.ipv4.tcp_wmem = 4096 32768 131072# /sbin/init 6...# /sbin/sysctl net.ipv4.tcp_wmemnet.ipv4.tcp_wmem = 4096 32768 13107Note: All commands are run as System "root" user.

View the original article here