aws library now supports multipart; initial design

doc/bugs/S3_upload_not_using_multipart/comment_9_74b2a392a537dde1c28089f1deed940c._comment

@@ -0,0 +1,31 @@
+[[!comment format=mdwn
+ username="joey"
+ subject="""comment 9"""
+ date="2014-10-28T16:42:21Z"
+ content="""
+The aws library now supports multipart uploads, using its
+S3.Commands.Multipart module.
+
+I don't think that multipart and chunking fit together: Typically the
+chunks are too small to need multipart for individual chunks. And the
+chunks shouldn't be combined together into a complete object at the end (at
+least not if we care about using chunking to obscure object size).
+Individual chunks sizes can vary when encryption is used, so combining them
+all into one file wouldn't work.
+
+Also, multipart uploads require at least 3 http calls, so there's no point
+using it for small objects, as it would only add overhead.
+
+So, multipart uploads should be used when not chunking, when the object to
+upload exceeds some size, which should probably defaut to something in the
+range of 100 mb to 1 gb.
+
+It might be possible to support resuming of interrupted multipart uploads.
+It seems that git-annex would need to store, locally, the UploadId,
+as well as the list of uploaded parts, including the Etag for the upload
+(which is needed when completing the multipart upload too).
+
+Also it should probably set Expires when initiating the multipart upload,
+so that incomplete ones get cleaned up after some period of time.
+Otherwise, users would probably be billed for them.
+"""]]