Rework CP and DP communication in IPv6 RD (VPP-1256),

https://gerrit.fd.io/r/#/c/12521/ 

introduced Java API performance regression.

Regression is visible when using packages from Nexus

(starting from v18.07-rc0~176-g5257452~b4253)

and when VPP is built locally.

Here are results for AclUpdateBenchmark with aclSize=1000:

Benchmark                      (aclSetSize)  (aclSize)  (mode)  Mode  Cnt   Score   Error  Units

AclUpdateBenchmark.testMethod           100       1000      L3  avgt  100  13.527 ± 0.341  ms/op

There might be some issues in the benchmark (work in progress), but results for 18.07-rc0~175-g17e5cfb~b4252 are noticeably better:

Benchmark                      (aclSetSize)  (aclSize)  (mode)  Mode  Cnt   Score   Error  Units

AclUpdateBenchmark.testMethod           100       1000      L3  avgt  100  7.196 ± 0.446 ms/op

For smaller aclSize=100 difference is even higher:

0.840 ± 0.020 before, but 11.129 ± 0.329 after,

for classify tables:

0.107 ± 0.003  ms/op before, but 11.648 ± 1.365 after.