Questions about read/write performance with replication enabled

[Pengfei Wu]

Hi There,

I understood the overhead of Seaweed is limited so that the read performance is (almost) the same as the underlying HDD/SSD performance for a single node setup. I am wondering with Filer fuse mount, will Seaweed concurrently read from multiple replicas with sub-linear performance boost (limited by the network) for a single reader? And does Seaweed (or Filer fuse mount) have any load balance algorithm for multiple concurrent readers? 

When a volume is in readonly mode (with the failure of disk or node), will Filer fuse mount simply complain io error to the writter? Is there any way to detect which files are in ro mode so that the upper level system can prevent writting instead of getting io error while doing so?

Thanks in advance!

BR,

Pengfei

[Chris Lu]

For a single reader, currently "weed mount" only sequentially reads the file content.

The reads are randomized for replicated data.

If one volume server is not writable,  the "weed mount" will ask the master to assign a different file id, which usually goes to a different volume server.

--
You received this message because you are subscribed to the Google Groups "Seaweed File System" group.
To unsubscribe from this group and stop receiving emails from it, send an email to seaweedfs+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/seaweedfs/39c8b5cc-18d6-4536-9d73-2b2041303fa6n%40googlegroups.com.