Hi, 2005/5/11, Emmanuel Lecharny <[EMAIL PROTECTED]>: > > > I have a question about the ProtocolCodec stuff : > > > - is it possible to have an encoder that can push chunks of PDU > instead > > > of pushing a whole PDU ? > > > > You can push chunks of PDU by calling write(ByteBuffer) multiple > > times. But MINA will flush the list of buffers you pushed in after > > encode() method returns. > > Isn't it possible that MINA flush the chunks before the encode method > return ?
I can fix it to do so. > > So You'll be able to limit the maximum size > > of ByteBuffer you allocate if you're using autoExpand mode, but MINA > > doesn't flush them until encode() method returns. (WDYT about > > autoExpand mode btw? ;) > > It would be very coll if neither the encoder nor MINA creates a 10Mb > ByteBuffer before sending it (think about a LdapSearch which returns all > Ldap entries ...) In this case, the data will be store twice : once in > the POJO and once in the ByteBuffer. not very good ... You're right, but I think MiNA can force users to use smaller buffers. But we'll have to write somewhere like FAQ about this issue. > on a muli-processor computer, it will make a better usage of each > processor... But I'm thinking of MINA as a front-end, like Apache is for > J2EE servers. may be it's a little bit too early ? > > Another advantage would be that MINA could send DNS requests to a > DNS/ApacheDS which run on a JVM, Kerberos requests to a > Kerberos/ApacheDS server, Ldap requests to a Ldap/ApacheDS server,.... > If the Ldap server is down, you still have the DNS and Kerberos servers > running. That means that maintenance operations are possible without > stopping all the services. Of course, this is only possible if MINA is > able to handle connexion on different ports (I don't know MINA enough to > tell). MINA could then be a front-end that dispatch requests to the > servers, so you don't have to launch an instance of MINA for each > server. Does it make sense ? Yes, it makes sense. But we'll have to consider the latency issues, too, though latency becomes zero if we use VM pipe. Trustin -- what we call human nature is actually human habit -- http://gleamynode.net/
