Ticket UUID: | 557488 | |||
Title: | add '-keepalive' option to tcp channel driver. | |||
Type: | Patch | Version: | None | |
Submitter: | davygrvy | Created on: | 2002-05-17 22:50:18 | |
Subsystem: | 27. Channel Types | Assigned To: | andreas_kupries | |
Priority: | 1 Zero | Severity: | ||
Status: | Closed | Last Modified: | 2004-10-27 02:49:20 | |
Resolution: | Rejected | Closed By: | andreas_kupries | |
Closed on: | 2004-10-26 19:49:20 | |||
Description: |
Here's a "-keepalive" option added to the windows tcp channel driver. Just windows is done. Is this worth following through and adding for all with docs? | |||
User Comments: |
davygrvy added on 2004-10-27 02:31:24:
Logged In: YES user_id=7549 no prob. there's a world of fun in all the setsockopt stuff. It's just too easy to implement ;) andreas_kupries added on 2004-10-27 02:02:11: Logged In: YES user_id=75003 Did a bit of talking on the chat. There it is considered not truly worth of implementation, but something better done in higher layers, i.e. application and protocol levels. Please reopen if there is some defending argument for this. [11:40]akuWhat are thoughts regarding 557488 ? In principle it requires a TIP, because a new fconfigure option is added. sockets only, windows only. What use has SO_KEEPALIVE ? Is it useful for us to have this ? [11:44]jenglishThe only places I've seen SO_KEEPALIVE used have been cargo-culted. It doesn't seem to be a useful thing to set. [11:45]akuOk, one indicator against. [11:45]jenglish http://www.unixguide.net/network/socketfaq/4.7.shtml explains the semantics. [11:45]CameronI've experimented with SO_KEEPALIVE. I don't know of any real *use* in contemporary applications. [11:48]akujoe - thanks ... reading ... [11:53]akuWell, keep-alive seems to be well-meaning, and sensible. OTOH I do not know of any app using it [11:54]akuThe only place where I heard the same term is for http, and there it just means that one conn is used for more than one request. [11:54]jenglishHTTP keep-alives are very different from SO_KEEPALIVE. [11:55]akuSee above, I ack this. It isjust the same term, but nothing on the socket-level [11:55]jenglishWell, not *very* different, the intent is the same; but HTTP keep-alives are done at the application level instead of in the transport layer. [11:55]akuright [11:56]jenglishAFAIK, the general consensus is that doing it at the application level is preferable, since you have more control. [11:57]akuOk. I will take this from the chat, add it to the report, lower the prio and then see if davidg responds in defense davygrvy added on 2002-11-27 09:32:30: Logged In: YES user_id=7549 was added to HEAD, but locked inside some comment blocks. davygrvy added on 2002-05-18 05:50:18: File Added - 23299: patch.txt |
Attachments:
- patch.txt [download] added by davygrvy on 2002-05-18 05:50:18. [details]