reduce buffer size of "unused" side of socket#4313
Conversation
gregcusack
force-pushed
the
reduce-net-buffer-sizes
branch
2 times, most recently
from
85a196a to
e62193f
Compare
gregcusack
requested review from
alexpyattaev and
steviez
and removed request for
steviez
steviez
left a comment
steviez
left a comment
There was a problem hiding this comment.
Ahh, this one slipped through the cracks. I doubt it will make any difference, but can you please rebase to tip as a precaution. I'm guessing your branch is on two-month-old-tip-of-master right now
gregcusack
force-pushed
the
reduce-net-buffer-sizes
branch
from
e62193f to
83a9a82
Compare
rebased! |
alexpyattaev
left a comment
alexpyattaev
left a comment
There was a problem hiding this comment.
testing on unstaked mnb node does not reveal any adverse impacts. will start on testnet, if all goes well I'll merge it in the morning.
alexpyattaev
left a comment
alexpyattaev
left a comment
There was a problem hiding this comment.
ok tested on staked node, TPU works fine, LGTM!
gregcusack
force-pushed
the
reduce-net-buffer-sizes
branch
from
83a9a82 to
14de739
Compare
| udp_config, | ||
| quic_config, | ||
| tpu_config, | ||
| tpu_reuseport_config, |
There was a problem hiding this comment.
This is actually wrong, quic port should not use REUSEPORT variant. I suggest we merge this PR after #5832 , it will also become much smaller.
There was a problem hiding this comment.
oooh the current code is using reuseport. why do we not want to use reuseport here? and ya no prob waiting.
There was a problem hiding this comment.
because it allows it to bind on top of an already occupied port here =)
|
Clearing out the inbox - are we still interested in pursuing this one @gregcusack ? |
ahh yes we are. haven't gotten a chance to come back to this after recent refactor of all the socket libraries. we can close and i can create a new PR |
3 participants
Follow Up to PR: #3929
Problem
We allocate large send and receive buffers for all sockets. However, not all sockets read and write equally from/to their buffers. In fact, 6 sockets are primarily Read and 2 are primarily Write. Meaning we are allocating memory that is never used.
Summary of Changes
Services/Sockets
Read/Write
Gossip
RPC - TCP
Ip_echo - TCP
Repair
Repair_quic
Serve_repair
Serve_repair_quic
Ancestor_hashes_requests_quic
Ancestor_hashes_requests
Primarily Read
TVU
tvu_quic
Tpu
Tpu_forwards
Tpu_vote
Tpu_quic
Tpu_forwards_quic
Tpu_vote_quic
Primarily Write
Retransmitter
Broadcast
Follow Up PR: