I brought this up a bit in Mooloolaba. As a guess, I'm thinking that the pack-on-the-fly code may be slowing down the fetch. Even worse, it may be interacting poorly with the Nagle algorithm, so that once the pack has completed, the TCP auto-negotiation has decided that it shouldn't send as much content.
It is also possible that we have some server side issues, where we aren't keeping the write buffer full. However, I would suspect the source side before I suspect the target side.
In contrast I'm uploading a picture of network throughput for iterating over 'get_stream()' from Babune to my machine. Note that it is *completely* flat. Something is throttling, though, as it transferred at a steady 100kB/s when I should have at least 300kB/s of download bandwidth.
I'll try a few more tests, including checking to see what the throttling issue is, what 'branch' looks like from babune, and whether 'get_stream()' from launchpad is also as flat.
I brought this up a bit in Mooloolaba. As a guess, I'm thinking that the pack-on-the-fly code may be slowing down the fetch. Even worse, it may be interacting poorly with the Nagle algorithm, so that once the pack has completed, the TCP auto-negotiation has decided that it shouldn't send as much content.
It is also possible that we have some server side issues, where we aren't keeping the write buffer full. However, I would suspect the source side before I suspect the target side.
In contrast I'm uploading a picture of network throughput for iterating over 'get_stream()' from Babune to my machine. Note that it is *completely* flat. Something is throttling, though, as it transferred at a steady 100kB/s when I should have at least 300kB/s of download bandwidth.
I'll try a few more tests, including checking to see what the throttling issue is, what 'branch' looks like from babune, and whether 'get_stream()' from launchpad is also as flat.