Initial Block Download optimisations #5099
Conversation
rebroad
force-pushed
the
headersfirst8-improvements
branch
2 times, most recently
from
cf739b6 to
410dbef
Compare
|
There are probably some commits in here that can be removed. I've kept it fairly granular so that some of the changes are easily optional. The big change is the commit called "Track partial download progress...", and I've recently updated this so that it now changes the window size per node based upon the bandwidth each note is achieving. @laanwj Is it worth tagging members of the core development team in a comment to draw their attention to this? |
|
@rebroad Possibly. But it could be that this list of change is too intimidating for one pull, and it may be better to trickle the changes instead of having one umbrella "Headersfirst8 improvements". |
rebroad
force-pushed
the
headersfirst8-improvements
branch
19 times, most recently
from
e56b9a7 to
7ba63dd
Compare
rebroad
changed the title
rebroad
force-pushed
the
headersfirst8-improvements
branch
3 times, most recently
from
ebb170f to
854367d
Compare
|
tumbleweed |
rebroad
force-pushed
the
headersfirst8-improvements
branch
from
854367d to
86de964
Compare
rebroad
force-pushed
the
headersfirst8-improvements
branch
6 times, most recently
from
2c6c439 to
585e89c
Compare
… disconnect logic for stalls.
rebroad
force-pushed
the
headersfirst8-improvements
branch
from
585e89c to
ee457ac
Compare
|
I don't know what travis is talking about - it seems to be running an old version that isn't present in the current branch. |
rebroad
force-pushed
the
headersfirst8-improvements
branch
from
ee457ac to
e50e834
Compare
rebroad
force-pushed
the
headersfirst8-improvements
branch
from
e50e834 to
96e8bf5
Compare
…cond. This speeds up IDB.
rebroad
force-pushed
the
headersfirst8-improvements
branch
from
96e8bf5 to
e2b98a9
Compare
|
Needs rebase. |
2 participants
I've been developing these improvements in parallel with headersfirst8.
It's now been optimised for use with early to late block chain so adapts to the block sizes and also the bandwidth of each node to make optimal use of the connections available.
The sipa disconnect code has been disabled as it usually disconnected the most important node (with my changes), but it will still display when the logic would have executed.
Please do test and compare download times with before this test and I'm sure you'll see improvements.
It's not yet optimised for use with earlier blocks (which were small enough to download in batches of 500 each), but that will be coming shortly. Raising the pull request so that people can view and comment on some of the changes.