diff options
Diffstat (limited to 'cpp')
| -rw-r--r-- | cpp/demo/Ice/throughput/README | 8 |
1 files changed, 7 insertions, 1 deletions
diff --git a/cpp/demo/Ice/throughput/README b/cpp/demo/Ice/throughput/README index d4b56c0b205..df67f6fde96 100644 --- a/cpp/demo/Ice/throughput/README +++ b/cpp/demo/Ice/throughput/README @@ -16,7 +16,13 @@ which on the server side means that the operation implementation is passed pointers into the marshalling buffer which eliminates a copy of the sequence data. -For return aramaters we also use the cpp:array" metadata, but this +For return paramaters we also use the cpp:array" metadata, but this time in conjuction with AMD. Doing this also reduces the number of times the returned sequence is copied during marshalling. +The performance for byte sequences is expected to be greater than +for other types, because the cost of marshaling and unmarshaling +is greater for more complex types, with byte sequences there isn't +need to allocate/dellocate objects witch could be expesive, the C++ +byte seqences mapping also uses a zero copy optimization which is +not possible with other types. |