Comparison of Fastdfs and GFs and NFS

Because I have just been involved in the field of distributed file storage, recently read a lot of documents, now feel a little dizzy, currently interested in Fastdfs, want to introduce a project, to ask, this system and GFS and NFS contrast, what is the difference and advantages. These two may not be a category, perhaps with kfs,gluster or lustre closer to some, can describe the difference, because the document is very few, such a good thing to promote, but also have a more clear comparison and explanation, it is easier to do a solution to convince the boss. ^_^
For example, I have my own Apache server, and then back-end access to GFS or NFS, and the use of Fastdfs method is no different. The pros and cons are. Read the Facebook program, it seems that the app Server with NFS file storage, but do not understand these, please guide the owner, said the wrong place please forgive me ...

=================================================================================

I suggest you read more about the posts in the Forum. Fastdfs is a dedicated file system, similar to the mogilefs, needs to use a specialized API to access, he is not a common file system, can not be used in the form of path mount.

Special file system relative to the common file system, more concise and efficient, suitable for certain occasions, such as Web site storage pictures, video files and so on.

Note: Both NFS and lustre belong to the common file system. I don't know much about GFs, KFS and gluster, so I don't comment.
Please feel free to communicate.

=================================================================================

By looking at forum posts, I probably understand the difference and use of common file system, ask, Fastdfs is suitable for large capacity small file (less than 2m), high concurrent system use. If it is a small file, each access to query Tracker server, the cost will be too large, in addition to tracker server in this form, high concurrent pressure will not be too big. The problem is relatively junior, not too familiar with the distributed file, I hope to start talking about, such as the common picture server such an application environment, what advantages.

=================================================================================

Fastdfs's design goal is to support large capacity and high traffic. For a large number of small files, you can support very well.
Tracker server maintains storage server information in memory, query performance does not have any problems, just one more network transmission.

Fastdfs is suitable for storing files uploaded by users, such as user photos. If only the static files of the site (such as decorative pictures, CSS, JS, etc.), it is not necessary to use FASTDFS.

=================================================================================

Read the document, quite comprehensive, basically have already understood, very good, thank the author timely reply.

Now for a question to understand, the file name is generated by the system, then access to the file, you must use the volume name plus the system generated file name for access. So I have to maintain the mapping of file names, because in my system, the filename is a GUID, is not duplicated, there is no possibility to directly through this file name for distributed storage, I only passed in a GUID file name, you can access the file. For example, after two development to achieve this function.

=================================================================================

>> at present a question to understand, the file name is generated by the system, then access to the file, you must use the volume name plus the system generated file name for access.

Yes. The upload file is successful and Fastdfs returns a file ID (including the volume name and filename).

>> so I'm going to maintain the mapping of file names, because in my system, the file name is a GUID, will not repeat, there is no possibility to directly through this file name for distributed storage, I only passed in a GUID file name, you can access the file. For example, after two development to achieve this function.

The file identity is also not duplicated, and the GUID is a one by one corresponding relationship. It is recommended that you save the file identity and the GUID.

=================================================================================

I'll ask you a question, happy fish. You recommend how to save the file index information. I understand that this problem is related to the application environment, for example, I use Fastdfs to create a multi-user picture station, I worry about the number of files, index information maintenance and retrieval will become a bottleneck, if this index information is lost, but also a disaster.

=================================================================================

For user pictures, you can index by user. If the number of pictures per user is not too much, the index information can be stored in another open source project FASTDHT (distributed hash system) that I write.

Index this block, I'm going to write a distributed indexing system, which can increase, delete, change and check the index, and support the large index information of single object.

=================================================================================

I think it's better to keep the index information in the database.

Http://219.148.35.28/thread-1989559-1-14.html