d/freebsd-nq
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Use insertion sort instead of bubble sort in TCP LRO.

Browse Source 
Replacing the bubble sort with insertion sort gives an 80% reduction
in runtime on average, with randomized keys, for small partitions.

If the keys are pre-sorted, insertion sort runs in linear time, and
even if the keys are reversed, insertion sort is faster than bubble
sort, although not by much.

Update comment describing "tcp_lro_sort()" while at it.

Differential Revision:	https://reviews.freebsd.org/D6619
Sponsored by:	Mellanox Technologies
Tested by:	Netflix
Suggested by:	Pieter de Goeje <pieter@degoeje.nl>
Reviewed by:	ed, gallatin, gnn, transport
This commit is contained in:
Hans Petter Selasky 2016-06-03 08:35:07 +00:00
parent 8e7b04858c
commit ec6689059d
1 changed files with 8 additions and 11 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

19
sys/netinet/tcp_lro.c
View File

@ -387,7 +387,8 @@ tcp_lro_msb_64(uint64_t x)
* number of elements to sort and 64 is the number of sequence bits
* available. The algorithm is bit-slicing the 64-bit sequence number,
* sorting one bit at a time from the most significant bit until the
* least significant one, skipping the constant bits.
* least significant one, skipping the constant bits. This is
* typically called a radix sort.
*/
static void
tcp_lro_sort(struct lro_mbuf_sort *parray, uint32_t size)
@ -399,17 +400,13 @@ tcp_lro_sort(struct lro_mbuf_sort *parray, uint32_t size)
uint32_t y;
repeat:
/* for small arrays bubble sort is faster */
/* for small arrays insertion sort is faster */
if (size <= 12) {
for (x = 0; x != size; x++) {
for (y = x + 1; y != size; y++) {
if (parray[x].seq > parray[y].seq) {
/* swap entries */
temp = parray[x];
parray[x] = parray[y];
parray[y] = temp;
}
}
for (x = 1; x < size; x++) {
temp = parray[x];
for (y = x; y > 0 && temp.seq < parray[y - 1].seq; y--)
parray[y] = parray[y - 1];
parray[y] = temp;
}
return;
}