Change Details

A test on lip5: Recv Send Send Socket Socket Message Elapsed Size Size Size Time Throughput bytes bytes bytes secs. 10^6bits/sec 65536 32768 32768 30.00 4142.55

A test on lip5: I used the counting patch of D35390 to add 2 new sysctls. One counts nodes traversed in the tree. The other accumulates tree sizes. I ran 4 MAERTS tests, and recorded the effects on the stats of each test. Then I merged the counting patch of D35390 and this one (which requires a bit of hand-manipulation of a couple of loops) and applied the same test to the resulting kernel. For the base case: ``` # lookups size change 75184049 61872494055 79019334 63448441073 75385477 60707783212 ``` For the modified code: ``` # lookups size change 42367978 66462958352 45377035 65421422630 46444115 65362449399 ``` So, it seems there's some benefit.

A test on lip5: Recv Send Send Socket Socket Message Elapsed Size Size Size Time Throughput bytes bytes bytes secs. 10^6bits/sec I used the counting patch of D35390 to add 2 new sysctls. One counts nodes traversed in the tree. The other accumulates tree sizes. I ran 4 MAERTS tests, and recorded the effects on the stats of each test. Then I merged the counting patch of D35390 and this one (which requires a bit of hand-manipulation of a couple of loops) and applied the same test to the resulting kernel. 65536 32768 32768 30.00 4142.55 For the base case: ``` # lookups size change 75184049 61872494055 79019334 63448441073 75385477 60707783212 ``` For the modified code: ``` # lookups size change 42367978 66462958352 45377035 65421422630 46444115 65362449399 ``` So, it seems there's some benefit.