CentOS 5 kernel has read the default buffer size of TCP, default: net.ipv4.tcp_rmem = 4096 87380 4194304
The solution is the last digit piecemeal point, the specific operation is to add a line to the file /etc/sysctl.conf:
net.ipv4.tcp_rmem = 4096 87380 174760
Then save
Restart the network service network restart, OK, or if some sites do not, you can check whether the /etc/sysctl.conf file and following the same
net.ipv4.ip_local_port_range = 1024 65536
net.core.rmem_max=174760
net.core.wmem_max=16777216
net.ipv4.tcp_rmem=4096 87380 174760
net.ipv4.tcp_wmem=4096 65536 16777216
net.ipv4.tcp_fin_timeout = 15
net.ipv4.tcp_keepalive_time = 600
net.ipv4.tcp_tw_recycle = 1
net.core.netdev_max_backlog = 30000
net.ipv4.tcp_no_metrics_save=1
net.core.somaxconn = 262144
net.ipv4.tcp_syncookies = 1
net.ipv4.tcp_max_orphans = 8000
net.ipv4.tcp_max_syn_backlog = 8000
net.ipv4.tcp_synack_retries = 2
net.ipv4.tcp_syn_retries = 2
net.ipv4.tcp_wmem = 4096 65536 16777216: memory for each socket used to automatically tune definition. The first value 4096 is the minimum number of bytes to transmit socket buffer allocation. 65536 second value is the default value (which is covered wmem_default), the buffer system load is not heavy to the case where this value can grow. 16777216 third value is the maximum number of bytes in the transmit buffer space (this value is overwritten wmem_max)
net.ipv4.tcp_rmem = 4096 87380 174760: tcp_wmem with similar, but it represents the value of the reception buffer is used for automatic tuning.
net.core.rmem_max = 25165824 # defines the maximum TCP / IP stack of the receive window size
net.core.rmem_default = 25165824 # define a default TCP / IP stack of the receive window size
net.core.wmem_max = 25165824 # send window defines the maximum size of TCP / IP stack
net.core.wmem_default = 65536 # define a default TCP / IP stack in the send window size
net.ipv4.tcp_sack = 1 # Enable selective response (Selective Acknowledgment), which may be out of order by the response received packets selectively to improve performance (which would allow the sender to send only the missing segment); (for wide-area network for) this option should be enabled, but this will increase the CPU occupancy.
net.ipv4.tcp_window_scaling = 1 # defined in RFC1323 is enabled, supports more than 64K window
net.ipv4.tcp_fack = 1 # Enable forwarding response (Forward Acknowledgment), which can be selectively response (SACK) to reduce congestion situation; this option should also be enabled.
net.ipv4.tcp_mem 24576 32768 49152 determine how the TCP stack should reflect memory usage; each unit value is memory pages (usually 4KB). The first value is the lower limit of memory usage. The second value is the upper limit of the memory buffer pressure mode start using the application of pressure. The third value is the memory limit. The message can be discarded at this level, so as to reduce the use of memory. For larger BDP can increase these values (but remember, the unit is the memory pages, not bytes).