dev@grizzly.java.net

Re: Is Java 7 NIO2 API incooreprated into the newest Grizzly 2.3.11

From: Oleksiy Stashok <oleksiy.stashok_at_oracle.com>
Date: Wed, 12 Mar 2014 18:16:30 -0700

Hi Ming Qin,

On 12.03.14 11:54, Ming Qin wrote:
>
> To find a solution for this issue you described on Linux ubuntu kernel
> version 2.6.32-21 generic #21 and OpenJDK 7 source codes, would you
> consider that is feasible approach - investigation and modification
> OpenJDK JNI source codes interacting with Native Posix Thread
> Liberary's epoll or select(/poll) loops?
I think the issue is mainly related to Java part of NIO.2 API, for sure
it may involve JNI part as well, but it's different story.

Thanks.

WBR,
Alexey.



>
> Ming Qin - mingqin.wordpress.com
> Cell Phone 949-388-9898
>
>
> On Monday, March 10, 2014 12:46 AM, Oleksiy Stashok
> <oleksiy.stashok_at_oracle.com> wrote:
> Hi Gregory,
>
>> Could you elaborate the drawbacks?
> the one for which I can't a workaround is that you have to allocate
> ByteBuffer for asynchronous read(...) operation. At the time you
> allocate this ByteBuffer you don't know if there's data available for
> reading or not (unlike NIO.1, where first you get an event, that
> there's some data to be read and only then you allocate a ByteBuffer).
> IMO this drawback can potentially lead to DoS attacks.
>
> Makes sense?
>
> Thanks.
>
> WBR,
> Alexey.
>
>>
>> On Mar 7, 2014, at 13:41, Oleksiy Stashok <oleksiy.stashok_at_oracle.com
>> <mailto:oleksiy.stashok_at_oracle.com>> wrote:
>>
> Hi Ming Qin,
>
> On 07.03.14 06:46, Ming Qin wrote:
>> Hi Everyone:
>>
>> On the page of https://grizzly.java.net/transportsconnections.html,
>> there is a statement likes"Grizzly 2.3 has two Transport
>> implementations: NIO TCP and NIO UDP, which is based on Java NIO,
>> though in the future, support for NIO.2 based transports and SCTP
>> will be added once JDK 7 is available."
>>
>> How does Grizzly switch back and forth between NIO and NIO2
>> implementations ?
> for now we're not planning to implement NIO2 based Grizzly Transport,
> because of certain NIO2 drawbacks, SCTP is on our list, but at the
> moment we just don't have time for it (for sure contributions are
> welcome).
>
> Thanks.
>
> WBR,
> Alexey.
>
>
>
>