发信人: babycry (babycry), 信区: Database
标 题: Re: question on large tables (>=800 million records, 10 G b
发信站: BBS 未名空间站 (Tue Jan 23 12:04:39 2007)
Can you give us the reason why you would consider BDB ?
I would consider BDB as the worse choice. Here is my reason:
According to the features provided by BDB,
* It provides transactional tables.
Transaction is not need at all for me for data mining.
What is more, I need to pay the price for having transactions,
and I cannot disable transaction feature.
* Each table has to have a primary key, which is stored with the table data
(to make things worse).
This means I need to have more I/O for each query, and am going to be even
* sequential scanning for BDB is slower.
Using index is slower for large tables, and I have to do table scan.
And sequential scanning for BDB is slower than other storage engines (come
* key values are not prefix or suffix compressed.
* there are holes in the table, and the tables using BDB format is larger
than MyISAM .
I do not need to enumerate more.
It seems like every single feature is the opposite of what I need.
【 在 watchman (飞来飞去~~第二十六式~~斜飞式) 的大作中提到: 】
: I would consider berkley DB.
※ 来源:·BBS 未名空间站 http://mitbbs.com·[FROM: 18.51.]