Excessive LAG on OGG extract Data Pump Sending Data Over WAN, calculate TCP TCPBUFSIZE TCPFLUSHBYTES [ID 1071892.1]

Version 10.4.0.12 and later

症状：

datapump进程严重lag，并未达到预期的网络吞吐量

诱因：

OGG的tcpbufsize默认30000字节，对繁忙网络设置过小；

系统没有启用tcp window scale；

系统的socket buffer过小；

诊断：

先通过ping查看RTT，并计算实际吞吐量；

开启tcp window scale；

提升socket buffer；

$ ping -s target 56 4

PING target: 56 data bytes

64 bytes from target.mydomain.com (10.221.65.15): icmp_seq=0. time=39. ms

64 bytes from target.mydomain.com (10.221.65.15): icmp_seq=1. time=39. ms

64 bytes from target.mydomain.com (10.221.65.15): icmp_seq=2. time=39. ms

64 bytes from target.mydomain.com (10.221.65.15): icmp_seq=3. time=39. ms

----redsap58 PING Statistics----

4 packets transmitted, 4 packets received, 0% packet loss

round-trip (ms) min/avg/max = 39/39/39

由上可得rtt=39ms，假定带宽为155Mbps，网络吞吐量=min(155M,socket_buffer/rtt)，则socket_buffer=155M*rtt/8=755625 bytes

将tcpbufsize/tcpflushbytes设置为755625

RMTHOST target, MGRPORT 7809, TCPBUFSIZE 755625, TCPFLUSHBYTES 755625；

开启tcp scale window;

增大socket buffer

net.core.rmem_max

net.core.wmem_max

net.ipv4.tcp_rmem

net.ipv4.tcp_wmem

net.ipv4.tcp_window_scaling

具体可参照

http://space.itpub.net/15480802/viewspace-757720

ITPub博客

goldengate的datapump调优