You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@mina.apache.org by Emmanuel Lécharny <el...@gmail.com> on 2013/03/22 17:54:51 UTC
[MINA3] Idle processing
Hi,
as I'm implementing the UDP server, and got really low performances
compared to the same test done on MINA2, I did some profiling. It
appears that the idle processing is a bit of a bootleneck, the way it's
currently implemented.
For instance, we grab the WRITE_IDLE_INDEX from the session attributes,
and update it :
public void sessionWritten(final AbstractIoSession session, final
long timeInMs) {
final Integer oldIndex = session.getAttribute(WRITE_IDLE_INDEX);
...
session.setAttribute(WRITE_IDLE_INDEX, index);
}
Accessing the WRITE_IDLE_INDEX attribute is done with this method :
public final <T> T getAttribute(final AttributeKey<T> key) {
return attributes.getAttribute(key);
}
which calls :
public <T> T getAttribute(AttributeKey<T> key) {
assertNotNull(key, "key");
readLock.lock();
try {
T value = (T) attributes.get(key);
return value;
} finally {
readLock.unlock();
}
}
So far, so good, except that we acquire a lock and release it for every
call.
Wouldn't it be better to store the WRITE_IDLE_INDEX as an AtomicLong in
the session itself ?
--
Regards,
Cordialement,
Emmanuel Lécharny
www.iktek.com
Re: [MINA3] Idle processing
Posted by Julien Vermillard <jv...@gmail.com>.
In DefaultAttributeContainer
private final Map<AttributeKey<?>, Object> attributes = new
HashMap<AttributeKey<?>, Object>();
we could use a ConcurrentHashMap, and get rid of the locks no ?
On Fri, Mar 22, 2013 at 8:57 PM, Julien Vermillard <jv...@gmail.com>wrote:
> hmm we have a lock on all session attributes, why ?
>
>
> On Fri, Mar 22, 2013 at 5:54 PM, Emmanuel Lécharny <el...@gmail.com>wrote:
>
>> Hi,
>>
>> as I'm implementing the UDP server, and got really low performances
>> compared to the same test done on MINA2, I did some profiling. It
>> appears that the idle processing is a bit of a bootleneck, the way it's
>> currently implemented.
>>
>> For instance, we grab the WRITE_IDLE_INDEX from the session attributes,
>> and update it :
>>
>> public void sessionWritten(final AbstractIoSession session, final
>> long timeInMs) {
>> final Integer oldIndex = session.getAttribute(WRITE_IDLE_INDEX);
>> ...
>> session.setAttribute(WRITE_IDLE_INDEX, index);
>> }
>>
>> Accessing the WRITE_IDLE_INDEX attribute is done with this method :
>>
>> public final <T> T getAttribute(final AttributeKey<T> key) {
>> return attributes.getAttribute(key);
>> }
>>
>> which calls :
>>
>> public <T> T getAttribute(AttributeKey<T> key) {
>> assertNotNull(key, "key");
>> readLock.lock();
>>
>> try {
>> T value = (T) attributes.get(key);
>>
>> return value;
>> } finally {
>> readLock.unlock();
>> }
>> }
>>
>> So far, so good, except that we acquire a lock and release it for every
>> call.
>>
>> Wouldn't it be better to store the WRITE_IDLE_INDEX as an AtomicLong in
>> the session itself ?
>>
>>
>>
>> --
>> Regards,
>> Cordialement,
>> Emmanuel Lécharny
>> www.iktek.com
>>
>>
>
Re: [MINA3] Idle processing
Posted by Julien Vermillard <jv...@gmail.com>.
hmm we have a lock on all session attributes, why ?
On Fri, Mar 22, 2013 at 5:54 PM, Emmanuel Lécharny <el...@gmail.com>wrote:
> Hi,
>
> as I'm implementing the UDP server, and got really low performances
> compared to the same test done on MINA2, I did some profiling. It
> appears that the idle processing is a bit of a bootleneck, the way it's
> currently implemented.
>
> For instance, we grab the WRITE_IDLE_INDEX from the session attributes,
> and update it :
>
> public void sessionWritten(final AbstractIoSession session, final
> long timeInMs) {
> final Integer oldIndex = session.getAttribute(WRITE_IDLE_INDEX);
> ...
> session.setAttribute(WRITE_IDLE_INDEX, index);
> }
>
> Accessing the WRITE_IDLE_INDEX attribute is done with this method :
>
> public final <T> T getAttribute(final AttributeKey<T> key) {
> return attributes.getAttribute(key);
> }
>
> which calls :
>
> public <T> T getAttribute(AttributeKey<T> key) {
> assertNotNull(key, "key");
> readLock.lock();
>
> try {
> T value = (T) attributes.get(key);
>
> return value;
> } finally {
> readLock.unlock();
> }
> }
>
> So far, so good, except that we acquire a lock and release it for every
> call.
>
> Wouldn't it be better to store the WRITE_IDLE_INDEX as an AtomicLong in
> the session itself ?
>
>
>
> --
> Regards,
> Cordialement,
> Emmanuel Lécharny
> www.iktek.com
>
>