If it is an issue with Socket close timeouts, can we have all sockets created by the FtpServer to set SO_LINGER to zero or a small number? Basically, call Socket.setSolinger on every socket we/MINA creates. Do you think this is the same setting as the one that you pointed in the FAQ?
Regards, Sai Pullabhotla www.jMethods.com On Thu, Oct 8, 2009 at 6:53 AM, Niklas Gustavsson <nik...@protocol7.com> wrote: > On Thu, Oct 8, 2009 at 1:46 PM, <a...@safe-mail.net> wrote: >> Just googled it and it says: >> "file server running Windows Server 2003 with 1 GB of RAM can efficiently >> support approximately 100,000 remote concurrent file handles, regardless of >> the size of the files. If your users are likely to have more than 100,000 >> files open at a time, plan to split this load across two or more servers." >> >> So it's not likely it's the cause, as there is only 1 file per user, and >> there is less than 2000 users. > > The problem with file handles is not due to the file the user > downloads, but the sockets used by the server. Each socket consumes a > file handle, and the socket lives also after the user has disconnect > for a certain time. If too many users connect during this period, you > might run out of handles. The max allowed handles is a setting in the > OS. You can find some more info in our FAQ: > http://mina.apache.org/ftpserver/faq.html > >> Also as the ftp server runs fine when restarted, and it takes couple of days >> to reach this issue, I don't think it's the os causing this. > > I agree this points to some other issue, but let make sure. > >> niklas, can you give a private email to send the netstat report to(when I'll >> have it)? As I do not want to make it public. > > Send it to the address from which I send these emails. Also, can you > include a thread dump? > > /niklas >