Tibco EMS is a real Enterprise solution

EMS 8.0 was tested on RHEL 6.4 using 80 Java concurrent clients sending messages to queue non-stop with time measurement after 1000 messages. Hardware platform was HP ProLiant BL460c Gen8 enterprise class server with 32 Intel Xeon E5-2650 cores clocked at 2GHz with 256KB cache L2 per core and 20MB cache L3 total per 8 cores (64 threads), 64 GB RAM DDR3 1333 MHz, Smart Array P220i Storage Controller with Flash Backed Write Cache (with battery backup), 10Gbps Ethernet.

EMS was tested using locally installed Oracle 12c database as a db storage and local GFS2 volume replicated over DRBD to another node - used as a file storage. Test were performed from host of passive node against active node with TCP/IP packet Round Trip Time 0.1/3ms and 0/10ms for DRBD inter-DC setup.

Test results resemble 0-1 digital signal what can be explained by allocation of new extents for Oracle DB tablespace.

Added RTT 3ms doesn't have visible influence on results so it can be understood that DB operations are major time component.

Visible jitter can be explained as internal overhead of GFS2.

Visible outlier is a result of changing RTT to default (0.1/0.1 ms) and then back to 3 / 10 ms.

EMS FS results can be improved via tuning deadline I/O scheduler and disabling storage flushes and barriers (FBWC has a battery!).

HornetQ 2.3 results have hardly visible sin^2 trend - probably thanks to JNI and kernel AIO.

Results may contain "grow and cut" GC pattern.

HornetQ results heavily depend on network performance so there is a room for ACK protocol optimization.