-
Notifications
You must be signed in to change notification settings - Fork 435
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
UCP/EP: Add eager multi-fragment overhead #7967
Conversation
e7b8cfc
to
07791cf
Compare
|
||
if (eager) { | ||
params->bw += 1.0 / ((1.0 / bw) + (iface_attr->overhead / | ||
iface_attr->cap.am.max_bcopy)); |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
is it possible max_bcopy would be 0?
src/ucp/core/ucp_ep.c
Outdated
bw = ucp_tl_iface_bandwidth(context, &iface_attr->bandwidth); | ||
|
||
if (eager) { | ||
params->bw += 1.0 / ((1.0 / bw) + (iface_attr->overhead / |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
pls add comment:
/* eager protocol has overhead for each fragment */
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
use
overhead = send_post_overhead + send_pre_overhead (from estimate_perf)
@@ -1749,7 +1751,14 @@ static void ucp_ep_config_calc_params(ucp_worker_h worker, | |||
} | |||
} | |||
|
|||
params->bw += ucp_tl_iface_bandwidth(context, &iface_attr->bandwidth); | |||
bw = ucp_tl_iface_bandwidth(context, &iface_attr->bandwidth); | |||
|
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
can remove space line
@@ -360,13 +360,21 @@ void test_ucp_peer_failure::do_test(size_t msg_size, int pre_msg_count, | |||
EXPECT_NE(UCS_OK, m_err_status); | |||
|
|||
if (UCS_PTR_IS_PTR(sreq)) { | |||
/* The request may either succeed or fail, even though the data is | |||
ucs_status_t status; | |||
/* TODO update comment |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
- why needed? if m_err_count!=0, i would expect the request to also complete
- update comment
- can we use request_wait()?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
discussed offline:
- use request_wait
- update comment: "If rendezvous protocol is used, the m_err_count is increased on the receiver side, so the send request may not complete immediately"
df01e95
to
987b782
Compare
src/ucp/core/ucp_ep.c
Outdated
status = uct_iface_estimate_perf(wiface->iface, &perf_attr); | ||
ucs_assert(status == UCS_OK); |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
fail config_init
@@ -1812,17 +1826,23 @@ static size_t ucp_ep_config_calc_rndv_thresh(ucp_worker_t *worker, | |||
(2 * eager_zcopy.overhead) + rndv.overhead) - | |||
eager_zcopy.reg_overhead - eager_zcopy.overhead; | |||
|
|||
denumerator = eager_zcopy.reg_growth + | |||
denominator = eager_zcopy.reg_growth + |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
yeah..
@Artemy-Mellanox can you pls squash? |
f30107e
to
68499ec
Compare
No description provided.