ipng/lcpng
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Allow larger fraction of CPU to be used by netlink

Browse Source 
Instead of doing BATCH_DELAY_MS work and BATCH_DELAY_MS sleep, add
a BATCH_WORK_MS (40) and lower BATCH_DELAY_MS (10), so we'll work
80% of the time, and consume netlink messages 20% of the time.

Also raise the total batch size to 8K because on my test machine
we run 2K in 13ms or 8K in ~50ms.
This commit is contained in:
Pim van Pelt
2021-08-29 17:47:33 +02:00
parent 7c864ed099
commit 6d86d2b075
2 changed files with 9 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
lcpng_netlink.c
View File

@ -47,6 +47,7 @@ lcp_nl_main_t lcp_nl_main = {
.rx_buf_size = NL_RX_BUF_SIZE_DEF,
.tx_buf_size = NL_TX_BUF_SIZE_DEF,
.batch_size = NL_BATCH_SIZE_DEF,
.batch_work_ms = NL_BATCH_WORK_MS_DEF,
.batch_delay_ms = NL_BATCH_DELAY_MS_DEF,
};
@ -255,7 +256,7 @@ lcp_nl_process_msgs (void)
lcpm->lcp_sync = 0;
/* process a batch of messages. break if we hit our batch_size
* count limit or batch_delay_ms time limit.
* count limit or batch_work_ms time limit.
*
* We do this, because netlink messages will continue to be sourced
* by the kernel, and we need to periodically read them before they
@ -277,10 +278,10 @@ lcp_nl_process_msgs (void)
break;
}
usecs = (u64) (1e6 * (vlib_time_now (vlib_get_main ()) - start));
if (usecs >= 1e3 * nm->batch_delay_ms)
if (usecs >= 1e3 * nm->batch_work_ms)
{
NL_INFO ("process_msgs: batch_delay_ms %u reached, yielding",
nm->batch_delay_ms);
NL_INFO ("process_msgs: batch_work_ms %u reached, yielding",
nm->batch_work_ms);
break;
}
}