Is throughput in the mapreduce metrics in MB or Mb - hadoop

Question

After running TestDFSIO I got the following metrics:
2019-04-30 09:50:35,790 INFO fs.TestDFSIO: Date & time: Tue Apr 30 09:50:35 EDT 2019
2019-04-30 09:50:35,791 INFO fs.TestDFSIO: Number of files: 100
2019-04-30 09:50:35,791 INFO fs.TestDFSIO: Total MBytes processed: 10000
2019-04-30 09:50:35,791 INFO fs.TestDFSIO: Throughput mb/sec: 376.9
2019-04-30 09:50:35,791 INFO fs.TestDFSIO: Average IO rate mb/sec: 387.16
2019-04-30 09:50:35,791 INFO fs.TestDFSIO: IO rate std deviation: 60.42
2019-04-30 09:50:35,791 INFO fs.TestDFSIO: Test exec time sec: 115.21
Is Average IO rate mb/sec in MegaByte or Megabits?

Answer 1

TestDFSIO is a usefult tool but the only available documentation is in its source code.
By looking at the code for TestDFSIO.java it seems that the throughput is expressed in Mebibytes per seconds.
In the source code one can see how troughput is computed:
" Throughput mb/sec: " + df.format(toMB(size) / msToSecs(time)),
Function toMB() is the number of bytes divided by MEGA:
static float toMB(long bytes) {
return ((float)bytes)/MEGA;
}
which is in turn the constant 0x100000L, that is the integer 1048576=1024*1024.
From the code:
private static final long MEGA = ByteMultiple.MB.value();
and
enum ByteMultiple {
B(1L),
KB(0x400L),
MB(0x100000L),
GB(0x40000000L),
TB(0x10000000000L);
...
So the throughput should be expressed in mebibytes/sec (MiB/sec) and not in megabytes (MB).