support bitswap configurability #8268
Conversation
petar
force-pushed
the
feat/bitswap-config
branch
from
0e5a335
to
6b1c06b
Compare
| This number controls fairness and can very from 250Kb (very fair) to 10Mb (less fair, with more work | ||
| dedicated to peers who ask for more). Values below 250Kb could cause thrashing. |
There was a problem hiding this comment.
Were these experimental numbers you came up with?
There was a problem hiding this comment.
Yeah. They are more of an example to demonstrate the interaction between fairness and thrashing. The 250KB is derived theoretically using the observation that 65KB is one packet, so utilizing anything under 5 packets per send (and then switching to another peer) feels like it may cause thrashing. In the extreme 65Kb (1 packet per send) would likely be thrashing because the block preparation pipeline (which involves a context switch to disk IO) would start costing more than the send itself.
also, move toc to the top, as noted in #8268 (comment)
|
Applied suggested changes to |
also, move toc to the top, as noted in #8268 (comment)
Co-authored-by: Gus Eggert <[email protected]>
- extracts bitswap init to separate file and defines implicit defaults as consts - fixes a typo where BitswapTaskWorkerCount was 128 (instead of 8) and BitswapEngineBlockstoreWorkerCount was 8 (instead of 128)
lidel
force-pushed
the
feat/bitswap-config
branch
from
37e430d
to
e05d94f
Compare
CI for go-ipfs-http-client fails but @aschmahmann is looking into it |
Part of #8233