|
|
Posted by Chung Leong on 05/01/07 01:14
On Apr 27, 5:02 am, Jerry Stuckle <jstuck...@attglobal.net> wrote:
> Using persistent connections mean you need to allocate the maximum
> connections you might ever need - and keep them allocated all the time.
> This takes resources away from both the OS and the DBMS (and your
> network if the DBMS is on a different server).
That's not entirely true. In a Apache + PHP system, there would be one
persistent connections (of one type) per child process. The number of
idle processes does not equal the maximum number of connections.
Apache will spawn more child processes as necessary. And it will shut
down extra spares if there are too many. Persistent connections owned
by these unwanted children would then be closed.
Persistent connections are problematic when the number of database
connections are limited. When the number of Apache processes exceeds
this number, obviously some of them wouldn't be able to get a
connection. And those that do own connections might not even be using
them--they could be handling file downloads for instance.
I think it's conceivable that a server could run out of database
connections while there is zero database activity:
1. A visitor goes to page to download a file
2. The PHP script open a persistent connection in order to log the
download in the database
3. The browser reuses the same HTTP connection (keep-alive) for the
download
4. For the next three hours, the child process handling the request
just sits on the DB connection
Repeat a hundred times or two and all connections will be exhausted.
Not an efficient scheme at all.
Navigation:
[Reply to this message]
|