tcmalloc
   cgroup
   cat /proc/123955/numa_maps

largepage
   HugeTLBfs
   oprofile

cat /proc/interrupts
irqbalance 智能均衡硬件中断
echo ff > /proc/irq/19/smp_affinity

cat /proc/softirqs

RPS: Receive Packet Steering 软件模拟的方式，实现了多队列网卡，分散多CPU上数据接收的负载。
RFS: Receive Flow Steering 把接收的数据包送达应用所在的CPU上，提高cache命中率。

IPX, Appletalk
注意观察dmesg有关TCP的抱怨
net.ipv4.tcp_fin_timeout = 32
net.ipv4.keeplive_time = 1200
...

提高初始拥塞窗口的大小（默认是3），大大减少短连接的响应时间。
TCP latency:
    3 -> 490ms
6 -> 475
10 -> 170
16 -> 465

磁盘的选择
文件系统
IO调度算法
pagecache的设置

libaic: Linux native aio，用eventfd做事件通知

CPU L1 cache                        0.5 ns
   CPU Branch mispredict               5   ns
   CPU L2 cache                        7   ns
   Mutex lock/unlock                   25  ns
   Main memory reference               100 ns
   Send 2K bytes over 1Gbps            20,000 ns
   Read 1MB sequentially from memory   250,000 ns
   Round trip within same datacenter   500,000 ns
   Disk seek (7200rpm)                 10,000,000 ns
   Read 1MB sequentially from disk     20,000,000 ns
   Send packet US->NL->US              150,000,000 ns

limited by slow data accessess to memory and disks

Intel Xeon 7400 CPU:
    96 KB L1 cache
16MB of L3
如何利用cache和空余的CPU计算力???
C = A <oper> B 实际上读取了一个block的A，一个block的B