2. The Two Component LSM-Tree Algorithm（2）

The operation of inserting an index entry into the memory resident C0 tree has no I/O cost.
插入数据到C0没有磁盘I/O操作。
However, the cost of memory capacity to house the C0 component is high compared to disk, and this imposes a limit on its size.
然而，C0组件占用更多内存，有大小的限制。
We need an efficient way to migrate entries out to the C1 tree that resides on the lower cost disk medium.
需要高效的把C0迁移到存储在磁盘上的C1的过程。
To achieve this, whenever the C0 tree as a result of an insert reaches a threshold size near the maximum allotted, an ongoing rolling merge process serves to delete some contiguous segment of entries from the C0 tree and merge it into the C1 tree on disk.
为了实现这一点，每当由于插入而导致的C0树达到接近分配的最大大小的阈值时，一个正在进行的滚动合并进程将从C0树中删除一些连续的条目片段，并将其合并到磁盘上的C1树中。（有道翻译）
Figure 2.2 depicts a conceptual picture of the rolling merge process.
图2.2描述了滚动合并过程的概念图。（有道翻译）

The C1 tree has a comparable directory structure to a B-tree, but is optimized for sequential disk access, with nodes 100% full, and sequences of single-page nodes on each level below the root packed together in contiguous multi-page disk blocks for efficient arm use;
C1树具有与b树类似的目录结构，但它对顺序磁盘访问进行了优化，节点100%满，根以下每一层的单页节点序列打包在连续的多页磁盘块中，以实现高效的arm使用;（有道翻译）
this opti- mization was also used in the SB-tree [21].
这种优化也被用于SB-tree[21]。（有道翻译）
Multi-page block I/O is used during the rolling merge and for long range retrievals, while single-page nodes are used for matching indexed finds to minimize buffering requirements.
多页块I/O用于滚动合并和长范围检索，而单页节点用于匹配索引发现以最小化缓冲需求。（有道翻译）
Multi-page block sizes of 256 KBytes are envisioned to contain nodes below the root; the root node is always a single page by definition.
256kbytes的多页块被设想包含根下的节点;根据定义，根节点始终是单个页面。（有道翻译）