It takes 3.5 seconds to upload a picture of 1,39 Mo. Just for information, I just tested Mirall 1.5 1.4.2 against OC 6 RC (realeased today).
![speed host userscloud vs filedrop speed host userscloud vs filedrop](https://staffmany.com/wp-content/uploads/2019/04/AdobeStock_166989896.jpeg)
(This is probably tricky because as far as I understand, the way csync handles transfers is limited to handling only a single file at a time.) do a single PROPFIND for a collection and evaluate all its members for changes from that one multistatus response (later, the PROPFIND depth could be increased even further, in theory up to requesting the complete remote hierarchy en bloc).Īlso, wouldn't it be possible to extend the owncloud implementation of the PUT command to support passing the modification time as a request header when uploading the file so that no additional PROPPATCH request is necessary after the upload?Īs suggested above, when actually transferring file data, batching several transfers in parallel could increase throughput. Maybe as a first step it would help to batch the metadata operations concerning the same collection, i.e. This also already affects walking the the remote filesystem when checking for modifications. I think the problem is primarily due to the overhead per file/collection (reading/setting props, caching the file locally etc.) which ends up being more costly than the actual data transfer for small files and collections.
![speed host userscloud vs filedrop speed host userscloud vs filedrop](https://easybuyou.guphotos.com/i/w?u=/images/O/9/OS0809/OS0809-1-cf82-8ZOg.jpg)
IMHO this also has to do with having a deep directory hierarchy. I can confirm this issue (OC 4.5.5, Linux client 1.1.4).