Tarsnap - What does the Tarsnap client do?

Consider a typical backup scenario:

In this example, we're looking at hourly backup #24584 of the entire hard disk of one of the tarsnap.com servers.

Read files: The client begins by reading data from the hard disk with code from libarchive.
In this particular example, there's 7588 MB on the disk.
Segment: The data from each file is split into context-dependent blocks or "chunks". The mathematics behind context-dependent chunkification is discussed in "From bsdtar to tarsnap: Building an online backup service."
After segmentation, we still have 7588 MB of data.
Deduplicate: The client recognizes chunks that it's already uploaded to the server. A simplified version is of this is explained in How does deduplication work?

Most of the data on disk has not changed since the previous archive, so we're left with 2.2 MB of new data.
Compress: Each block is compressed with the DEFLATE algorithm from zlib. Given the size of chunks produced by the segmentation stage, there is no significant difference in compression ratio between the DEFLATE and LZMA algorithms, but LZMA is significantly slower.
In our example, this produces 0.81 MB of compressed, new data.
Encrypt: Each compressed block is encrypted with your private keys so that nobody else can access your unencrypted data.
This does not add any noticeable amount of data, so we have 0.81 MB of encrypted, compressed, new data.
Upload: The "encoded" bytes (encrypted, compressed, new data) are uploaded to the Tarsnap service.