Instead of supporting full POSIX file system semantics, SeaweedFS choose to implement only a key~file mapping. Similar to the word "NoSQL", you can call it as "NoFS".
Instead of managing all file metadata in a central master, SeaweedFS choose to manages file volumes in the central master, and let volume servers manage files and the metadata. This relieves concurrency pressure from the central master and spreads file metadata into volume servers' memories, allowing faster file access with just one disk read operation!
SeaweedFS costs only 40 bytes disk storage for each file's metadata. It is so simple with O(1) disk read that you are welcome to challenge the performance with your actual use cases.
Here I will start one master node, and two volume nodes on port 8080 and 8081. Ideally, they should be started from different machines. Here I just use localhost as example.
Now you can save the fid, 3,01637037d6 in this case, to some database field.
The number 3 here, is a volume id. After the comma, it's one file key, 01, and a file cookie, 637037d6.
The volume id is an unsigned 32 bit integer. The file key is an unsigned 64bit integer. The file cookie is an unsigned 32bit integer, used to prevent URL guessing.
The file key and file cookie are both coded in hex. You can store the <volumeid,filekey,filecookie> tuple in your own format, or simply store the fid as string.
If stored as a string, in theory, you would need 8+1+16+8=33 bytes. A char(33) would be enough, if not more than enough, since most usage would not need 2^32 volumes.
If space is really a concern, you can store the file id in your own format. You would need one 4-byte integer for volume id, 8-byte long number for file key, 4-byte integer for file cookie. So 16 bytes are enough (more than enough).
(However, since usually there are not too many volume servers, and volumes does not move often, you can cache the results most of the time. Depends on the replication type, one volume can have multiple replica locations. Just randomly pick one location to read.)
Now when requesting a file key, an optional "dataCenter" parameter can limit the assigned volume to the specific data center. For example, this specifies that the assigned volume should be limited to 'dc1':
Usually distributed file system split each file into chunks, and a central master keeps a mapping of a filename and a chunk index to chunk handles, and also which chunks each chunk server has.
This has the draw back that the central master can not handle many small files efficiently, and since all read requests need to go through the chunk master, responses would be slow for many concurrent web users.
Instead of managing chunks, SeaweedFS choose to manage data volumes in the master server. Each data volume is size 32GB, and can hold a lot of files. And each storage node can has many data volumes. So the master node only needs to store the metadata about the volumes, which is fairly small amount of data and pretty static most of the time.
The actual file metadata is stored in each volume on volume servers. Since each volume server only manage metadata of files on its own disk, and only 16 bytes for each file, all file access can read file metadata just from memory and only needs one disk operation to actually read file data.
For comparison, consider that an xfs inode structure in Linux is 536 bytes.
### Master Server and Volume Server ###
The architecture is fairly simple. The actual data is stored in volumes on storage nodes. One volume server can have multiple volumes, and can both support read and write access with basic authentication.
All volumes are managed by a master server. The master server contains volume id to volume server mapping. This is fairly static information, and could be cached easily.
On each write request, the master server also generates a file key, which is a growing 64bit unsigned integer. Since the write requests are not as busy as read requests, one master server should be able to handle the concurrency well.
### Write and Read files ###
When a client sends a write request, the master server returns <volumeid,filekey,filecookie,volumenodeurl> for the file. The client then contact the volume node and POST the file content via REST.
When a client needs to read a file based on <volumeid,filekey,filecookie>, it can ask the master server by the <volumid> for the <volumenodeurl,volumenodepublicurl>, or from cache. Then the client can HTTP GET the content via REST, or just render the URL on web pages and let browsers to fetch the content.
Please see the example for details on write-read process.
In current implementation, each volume can be size of 8x2^32 bytes (32GiB). This is because of aligning contents to 8 bytes. We can be easily increased to 64G, or 128G, or more, by changing 2 lines of code, at the cost of some wasted padding space due to alignment.
Each individual file size is limited to the volume size.
### Saving memory ###
All file meta information on volume server is readable from memory without disk access. Each file just takes an 16-byte map entry of <64bitkey,32bitoffset,32bitsize>. Of course, each map entry has its own the space cost for the map. But usually the disk runs out before the memory does.
## Compared to Other File Systems##
Frankly, I don't use other distributed file systems too often. All seems more complicated than necessary. Please correct me if anything here is wrong.
Ceph can be setup similar to SeaweedFS as a key~blob store. It is much more complicated, with the need to support layers on top of it. Here is a more detailed comparison. https://github.com/chrislusf/seaweedfs/issues/120
SeaweedFS is meant to be fast and simple, both during usage and during setup. If you do not understand how it works when you reach here, we failed! Jokes aside, you should not need any consulting service for it.
SeaweedFS has a centralized master to lookup free volumes, while Ceph uses hashing to locate its objects. Having a centralized master makes it easy to code and manage. HDFS/GFS has the single name node for years. SeaweedFS now support multiple master nodes.
SeaweedFS can also store extra large files by splitting them into manageable data chunks, and store the file ids of the data chunks into a meta chunk. This is managed by "weed upload/download" tool, and the weed master or volume servers are agnostic about it.
Mongo's GridFS splits files into chunks and manage chunks in the central mongodb. For every read or write request, the database needs to query the metadata. It's OK if this is not a bottleneck yet, but for a lot of concurrent reads this unnecessary query could slow things down.
Since files are chunked(default to 256KB), there will be multiple metadata readings and multiple chunk readings, linear to the file size. One 2.56MB file would require at least 20 disk read requests.
On the contrary, SeaweedFS uses large file volume of 32G size to store lots of files, and only manages file volumes in the master server. Each volume manages file metadata themselves. So all the file metadata is spread onto the volume nodes memories, and just one disk read is needed.
More tools and documentation, on how to maintain and scale the system. For example, how to move volumes, automatically balancing data, how to grow volumes, how to check system status, etc.
This is a super exciting project! And I need helpers!
step 4: after you modify your code locally, you could start a local build by calling "go install" under $GOPATH/src/github.com/chrislusf/seaweedfs/go/weed
When testing read performance on SeaweedFS, it basically becomes performance test your hard drive's random read speed. Hard Drive usually get 100MB/s~200MB/s.
To modify or delete small files, SSD must delete a whole block at a time, and move content in existing blocks to a new block. SSD is fast when brand new, but will get fragmented over time and you have to garbage collect, compacting blocks. SeaweedFS is friendly to SSD since it is append-only. Deletion and compaction are done on volume level in the background, not slowing reading and not causing fragmentation.
