呢家啲server通常都喺行non uniform memory
啫係話一個cpu core/cluster同另一個cpu core/cluster可能唔會同一時間share memory (所以a記越來越多cpu core就開始浮現呢個問題)
如果唔係一喺lock一喺cpu cache做暗lock (參見MESI protocol)
代價就喺個cpu點都要等大家個memory/cache line同步 呢個時間可能係microsecond甚至millisecond
concurrent operation一多即刻爆slow query 進而導致好強烈嘅thrashing 變成惡性循環最終不得復生