1. Introduction（1）

As long-lived transactions in activity flow management systems become commercially available ([10], [11], [12], [20], [24], [27]), there will be increased need to provide indexed access to transactional log records.
随着长周期的事务管理系统商业应用，需要提供事务日志系统的索引操作。
Traditionally, transactional logging has focused on aborts and recovery, and has required the system to refer back to a relatively short-term history in normal processing with occasional transaction rollback, while recovery was performed using batched sequential reads.
传统事务日志关注挂掉和恢复，需要一个相对短时间记录批量读取恢复系统。
However, as systems take on responsibility for more complex activities, the duration and number of events that make up a single long-lived activity will increase to a point where there is sometimes a need to review past transactional steps in real time to remind users of what has been accomplished.
然而，随着系统负责的系统更加复杂，单个用户需要了解之前事物步骤，事件周期和数量要多的多。
At the same time, the total number of active events known to a system will increase to the point where memory-resident data structures now used to keep track of active logs are no longer feasible, notwithstanding the continuing decrease in memory cost to be expected.
随着系统总的事件数量增长，基于内存的日志系统在可预期情况下变得不可行了。
The need to answer queries about a vast number of past activity logs implies that indexed log access will become more and more important.
需要查询的数量巨大，日志查询系统的要求越来越重要。

Even with current transactional systems there is clear value in providing indexing to support queries on history tables with high insert volume.
即使现在的事务系统查询表格也是有很大量的插入操作。
Networking, electronic mail, and other nearly-transactional systems produce huge logs often to the detriment of their host systems.
网络应用、邮件和其他接近的事务系统产生巨大的日志损害主机系统。
To start from a concrete and well-known example, we explore a modified TPC-A benchmark in the following Examples 1.1 and 1.2.
一个众所周知的示例开始介绍，TPC-A。
Note that examples presented in this paper deal with specific numeric parametric values for ease of presentation;
注意示例使用了便于分析的特殊参数值。
it is a simple task to generalize these results.
这是一个简单任务生成结果。
Note too that although both history tables and logs involve time-series data, the index entries of the LSM-Tree are not assumed to have indentical temporal key order.
也要注意虽然历史表和涉及时间序列的数据，LSM-tree索引内容不以时间作为key。
The only assumption for improved efficiency is high update rates compared to retrieval rates.
唯一假设是高效的检索比较速度。