EZ2-VX6060
Software
SD.DT.F65-0e
Page
17
2.4
Linux Pre-Installed Benchmarks
Sample benchmarks are provided with the system. These benchmarks can help you get the current
performances of the system.
lmbench
covers most classical OS aspects,
iperf
will measure networking
performance on TCP/IP socket on ETHERNET. Using
iperf
on top of VXFabric (tm) is also possible when the
VXFabric(tm) software is installed.
2.4.1
What To Expect
VX6060 can run up to 8 execution threads on 2 CPUS. Communication between all the different threads can use
the following mecanism
1° between threads of same application: direct read/write into the common memory pool of the process : See
lmbench
bcopy
performance
2° between threads on same CPU in different processes:
>
Shared memory : see bcopy in lmbench
>
sockets: see iperf run locally with server and client on same node
3° between processes running on different CPUs
>
sockets: see iperf run between two nodes.
a. using simple GETH link between nodes
b. using a bound interface of two GETH links (see iperf chapter)
c. using VXFabric(tm) which gives sockets on top of the PCIE fabric between CPUA and CPUB
2.4.2
lmbench
lmbench is an old fashion benchmark which consists of a suite of simple, portable benchmarks. It is interesting
for results of memory bandwidth or latency.
To see the last
lmbench
result, change directory to /
opt/benchmarks/lmbench
and run the
make see
command.