正如有人已经提到的，“视情况而定”。如果数据库中的存储被认为是文件系统的1对1替代方案，那么它可能不是最佳选择。

但是，如果数据库后端将提供额外的值，而不仅仅是blob的序列化和存储，那么它可能是真正有意义的。

You may take a look at WKT Raster which is a project aiming at developing raster support in PostGIS which in turn serves as a geospatial extension for PostgreSQL database system. Idea behind the WKT Raster is not only to define a format for raster serialization and storage (using PostgreSQL system), but, what's much more important than storage, is to specify database-side efficient image processing accessible from SQL. Long story short, the idea is to move the operational weight from client to database backend, so it take places as close to storage itself as possible. The WKT Raster, as PostGIS, is dedicate to applications of specific domain, GIS.

要获得更完整的概述，请查看该系统的网站和演示文稿(PDF)。

I have worked with many digital storage systems and they all store digital objects on the file system. They tend to use a branch approach, so there will be an archive tree on the file system, often starting with year of entry e.g. 2009, subdirectory will be month e.g. 8 for August, next directory will be day e.g. 11 and sometimes they will use hour as well, the file will then be named with the records persistent ID. Using BLOBS has its advantages and I have heard of it being used often in the IT parts of the chemical industry for storing thousands or millions of photographs and diagrams. It can provide more granular security, a single method of backup, potentially better data integrity and improved inter media searching, Oracle has many features for this within the package they used to call Intermedia (I think it is called something else now). The file system can also have granular security provided through a system such as XACML or another XML type security object. See D Space of Fedora Object Store for examples.

有一件事我还没有看到任何人提到，但绝对值得注意的是，在大多数文件系统中也存在与存储大量图像相关的问题。例如，如果您采用上面提到的方法，以主键命名每个图像文件，在大多数文件系统上，如果您试图将所有图像放在一个大目录中，一旦您达到了非常大的图像数量(例如数十万或数百万)，您将遇到问题。

常见的解决方案是将它们散列到平衡的子目录树中。

我几乎从不把它们存储在数据库中。最好的方法通常是将映像存储在一个由中央配置变量控制的路径中，并根据DB表和主键(如果可能的话)命名映像。这给了你以下优势:

通过更新全局配置，将映像移动到另一个分区或服务器。 通过搜索图像的主键来查找与图像匹配的记录。 您的图像可以访问处理工具，如imagemagick。 在web应用程序中，您的图像可以由web服务器直接处理(节省处理)。 CMS工具和Coldfusion等网络语言可以处理本地上传。

如果您使用Teradata，那么Teradata Developer Exchange有一篇关于加载和检索lobs和blobs的详细文章。

http://developer.teradata.com/applications/articles/large-objects-part-1-loading

我们实现了一个文档成像系统，它将所有图像存储在SQL2005 blob字段中。目前有几百GB，我们看到了出色的响应时间和很少或没有性能下降。此外，fr法规遵从性，我们有一个中间件层，将新发布的文档归档到光学点唱机系统，该系统将它们公开为标准NTFS文件系统。

我们对结果非常满意，特别是在以下方面:

易于复制和备份 能够轻松实现文档版本控制系统