Comments on: Contrived FieldCache Load Test: Lucene 2.4 VS Lucene 2.9

By: Apache Lucene 2.9的改进 | 彭旭赣州seo优化

Apache Lucene 2.9的改进 | 彭旭赣州seo优化 — Sat, 25 Sep 2010 11:37:43 +0000

[...] Imagination的Mark Miller运行了一个简单的性能测试，表明在5,000,000个不同字符串下的情况下，Lucene [...]

By: 尝试使用IKVM运行Lucene 2.9.0版 | 彭旭赣州seo优化

尝试使用IKVM运行Lucene 2.9.0版 | 彭旭赣州seo优化 — Fri, 24 Sep 2010 15:49:55 +0000

[...] 　　在保证了正确性之后，要关注的便是性能了。根据我的推测，由于IKVM需要在Java生成的.NET程序集和BCL之间加上一层Runtime和JDK，因此其性能几乎一定会比Java原有的程序要差。不过，对于Lucene这种项目来说，算法才是性能的关键。例如，有人测试Lucene 2.9.0在某些情况下会比2.4有15倍左右的性能提升。不过由于没有很好的测试数据和场景，目前我只进行了最最简单的，不涉及磁盘IO的性能比较。 [...]

By: Mark Miller

Mark Miller — Thu, 24 Sep 2009 19:10:05 +0000

Partially – since I am just timing the loading of the FieldCache (and not doing it per segment). It’s just to make sure I have a bunch of segments – its only faster over multiple segments – its the same speed on an optimized Index. The more segments, the faster it is.

The reason that its also faster when you do it per segment (how Lucene works internally now), is that it avoids the speed trap that was in MultiTermEnum, and uses SegmentTermEnum – Yonik fixed that as well though, and this test shows the fruits of that. So essentially, it was both fixed and side stepped at the same time

By: Otis Gospodnetic

Otis Gospodnetic — Thu, 24 Sep 2009 18:15:09 +0000

mergeFactor=37 — presumably in order to avoid any segments to be merged during indexing, thus making it possible to show off the new and faster segment reloading.

By: Anonymous

Anonymous — Thu, 24 Sep 2009 03:46:18 +0000

Why did you use merge factor of 37?

By: Jim Murphy - PostRank

Jim Murphy - PostRank — Wed, 23 Sep 2009 00:16:49 +0000

This was our biggest issue by far. Its taken significant load off our servers when installing a new snapshot.

Thanks indeed Yonik.