Re: IOCP-Client question

Thanks for the links, I'll take a look.

Observational and empirical evidence is usually the most reliable
kind... ;)

Len

http://www.lenholgate.com

On Aug 27, 3:45 pm, Malachy Moses <malachy.mo...@xxxxxxxxx> wrote:
On Aug 27, 2:14 am, Len Holgate <len.holg...@xxxxxxxxxxx> wrote:

Also remember that even with ConnectEx, there is no way around the
Winsock stack's limitation of only up to 10 half-open connections.  In
other words, if you issue 20 quick ConnectEx's, the SYN for only the
first 10 will be transmitted.  The 11th SYN will be transmitted only
after SYN-ACK from one of the first 10 is received.- Hide quoted text -

Really, where is this documented?

Len

http://www.lenholgate.com

The limit of 10 on concurrent half-open connections is documented
under Part 2 of “Changes to Functionality in Windows XP Service Pack
2”  athttp://technet.microsoft.com/en-us/library/bb457156.aspx.
Under the section entitled "TCP/IP", scroll down to the sub-section
entitled "Limited number of simultaneous incomplete outbound TCP
connection attempts".

The limit of 10 was first introduced in WinXP SP2, and has been
maintained in Vista.  It was described by Microsoft as a "feature", to
"help to limit the speed at which malicious programs, such as viruses
and worms, spread to uninfected computers."  If you bump up against
the limit, event ID 4226 appears in the system event log.

There have been attempts to patch TCPIP.SYS to remove the limit.  The
most successful one I am aware of is found athttp://www.lvllord.de/,
but I believe it is available only for WinXP and not for Vista.

As for the precise algorithm by which the limit is achieved (i.e.,
send 10 SYNs but don't send the 11th until a first SYN-ACK is
received), I don't have documentary evidence.  It should be possible
to find it if you are interested, but I based my comments on
observational and empirical evidence.

.

Loading