@fighting-liu
similar results to above were observed running your example
```bash
>>> def mxnet_worker():
... b_time = time.time()
... import mxnet
... print 'time consumes: {}'.format(time.time()-b_time)
...
>>> read_process = [multiprocessing.Process(target=mxnet_worker) for i in
>>> range(8)]
>>> for p in read_process:
... p.daemon = True
... p.start()
...
>>> time consumes: 15.8834888935
time consumes: 15.884565115
time consumes: 15.8791670799
time consumes: 15.8853030205
time consumes: 15.8832161427
time consumes: 15.882764101
time consumes: 15.8819229603
time consumes: 15.8869299889
>>> read_process = [multiprocessing.Process(target=mxnet_worker) for i in
>>> range(8)]
>>> for p in read_process:
... p.daemon = True
... p.start()
...
>>> time consumes: 1.01575899124
time consumes: 1.02250099182
time consumes: 1.03319501877
time consumes: 1.03118515015
time consumes: 1.03451776505
time consumes: 1.03348302841
time consumes: 1.03426003456
time consumes: 1.03685307503
```
The second run gets the speed up again, I observed a similar trend in 1.1 as
well, although the gap is smaller.
I am investigating this issue further
[ Full content available at:
https://github.com/apache/incubator-mxnet/issues/12255 ]
This message was relayed via gitbox.apache.org for [email protected]
