京都内阁（TreeDB）的性能严重下降

Question

我选择 TreeDB 作为京都内阁后端，希望它能扩展到巨大的价值。不幸的是，有一个问题：

# ./kyotobench

Generated string length: 1024

1000 records, type t 74.008887ms throughput: 13511 /sec

2000 records, type t 145.390096ms throughput: 13756 /sec

4000 records, type t 290.13486ms throughput: 13786 /sec

8000 records, type t 584.46691ms throughput: 13687 /sec

16000 records, type t 1.150792756s throughput: 13903 /sec

32000 records, type t 2.134860729s throughput: 14989 /sec

64000 records, type t 4.378002268s throughput: 14618 /sec

128000 records, type t 9.41012632s throughput: 13602 /sec

256000 records, type t 20.457090225s throughput: 12513 /sec

512000 records, type t 45.934115353s throughput: 11146 /sec

1024000 records, type t 1m39.120917207s throughput: 10330 /sec

2048000 records, type t 3m41.720146906s throughput: 9236 /sec

4096000 records, type t 15m26.041653712s throughput: 4423 /sec

8192000 records, type t 5h5m31.431477812s throughput: 446 /sec

我打开一个 TreeDB，生成 2 个随机长度 ( 0<len<1024) 的随机字符串，并将它们分别用作键和值。代码：

http://pastebin.com/0HwHPXFq

这是什么原因？