v1.10: usnic fixes #933
v1.10: usnic fixes #933
Conversation
|
Test PASSed. |
|
Test PASSed. |
There was a problem hiding this comment.
So before this change, all sends were going over the priority channel (is that for acks?)
There was a problem hiding this comment.
Just small sends. See old line 1178 (a few lines above this one):
if (frag->sf_base.uf_type == OPAL_BTL_USNIC_FRAG_SMALL_SEND &&
frag->sf_ack_bytes_left < module->max_tiny_payload &&
// ...etc.
|
👍 |
|
@bturrubiates: Sorry, only this repo's organization members can interact with this bot. If you're a member of this organization, make your membership public so that this bot can verify your membership (go to http://github.com/orgs/open-mpi/people, find yourself on that page, then change your membership from "Private" to "Public"). |
|
Ben gave a 👍 |
|
Test PASSed. |
Messages should go on the data channel, even if they're short. Only ACKs go on the priority channel. (cherry picked from commit open-mpi/ompi@4de4a26)
Sequence numbers will wrap around; it is not sufficient to check for (seq-1) -- must use the SEQ_DIFF macro to properly handle the wraparound. This bug wasn't serious; it just meant we might retransmit one or two extra times when retransmits were triggerd and the sequence numbers wrapped around their sliding windows. (cherry picked from commit open-mpi/ompi@d624e0d)
Add endpoints in a blocked manner so that we don't overrun the fi_av_insert() event queue. Also make the AV EQ length an MCA param, and report it in mca_btl_base_verbose >=5 output. (cherry picked from commit open-mpi/ompi@db825ab)
(cherry picked from commit open-mpi/ompi@797d502)
A bunch of empirical testing has shown that increasing the retranmit timeout from 1ms to 5ms doesn't adversely affect performance, yet decreases the number of gratuitious retransmissions. (cherry picked from commit open-mpi/ompi@c2615a4)
jsquyres
force-pushed
the
pr/v1.10/usnic-fixes
branch
from
5e721ad to
2715e03
Compare
|
Test PASSed. |
Three minor updates from the code review of open-mpi/ompi-release#933: * Remove an extra blank line a show_help message * We no longer allow -1 for the MCA param btl_usnic_av_eq_num, so change the flag to REGINT_GE_ONE * Change "num_blocks" definition to be in terms of block_len (not eq_size)
Three minor updates from the code review of open-mpi#933: * Remove an extra blank line a show_help message * We no longer allow -1 for the MCA param btl_usnic_av_eq_num, so change the flag to REGINT_GE_ONE * Change "num_blocks" definition to be in terms of block_len (not eq_size) (cherry picked from commit open-mpi/ompi@9f3ed00)
Three minor updates from the code review of open-mpi#933: * Remove an extra blank line a show_help message * We no longer allow -1 for the MCA param btl_usnic_av_eq_num, so change the flag to REGINT_GE_ONE * Change "num_blocks" definition to be in terms of block_len (not eq_size) (cherry picked from commit open-mpi/ompi@9f3ed00)
|
Pushed one additional commit as a result from @bturrubiates' review. |
|
Test PASSed. |
|
@jsquyres any urgency to this? |
|
A release "sometime soon" would be nice -- I don't need an immediate release, however. Similar to other vendors, it's enough to know that it's actually merged in the community upstream. |
Three minor updates from the code review of open-mpi/ompi-release#933: * Remove an extra blank line a show_help message * We no longer allow -1 for the MCA param btl_usnic_av_eq_num, so change the flag to REGINT_GE_ONE * Change "num_blocks" definition to be in terms of block_len (not eq_size)
@bturrubiates Please review