Hi,
as I was looking at the NioSelectorProcessor, and more specifically the
part which reads data from the channel, I found that we are using a
global buffer (readBuffer). I have two concerns in this area :
- first, we may need to copy the buffer if we need to add an executor in
the chain. We may *not* have to do that, fait enough, but I'm quite sure
that we may encounter issue later if we don't. At least, I would make it
optionnal to protect us from such a potential issue. Am I alone having
this feeling ?
- second, I'm wondering if it would not be a better idea to either use a
wider buffer - say 64k -, or to iterate on the read until we have sucked
every bytes, storing them in as many buffers as needed, to avoid a
costly round trip through the chain for each 1024 bytes we will have in
the channel. Something like :
do {
int readCount = channel.read(readBuffer[bufNumber++]);
} while ( readCount >= 0);
There is some suggestion to make the allocated buffer configurable, and
I'm totally +1 for such an approach, but maybe starting with bigger
buffers, and eventually create other buffers to store the data, before
sending them into the chain, is also a good thing.
wdyt ?
--
Regards,
Cordialement,
Emmanuel Lécharny
www.iktek.com
