Send inn-workers mailing list submissions to
[email protected]
To subscribe or unsubscribe via the World Wide Web, visit
https://lists.isc.org/mailman/listinfo/inn-workers
or, via email, send a message with subject or body 'help' to
[email protected]
You can reach the person managing the list at
[email protected]
When replying, please edit your Subject line so it is more specific
than "Re: Contents of inn-workers digest..."
Today's Topics:
1. Re: nnrpd and TCP nagle (Julien ?LIE)
2. Re: nnrpd and TCP nagle
(The Doctor, 3328-138 Ave Edmonton AB T5Y 1M4, 669-2000, 473-4587)
----------------------------------------------------------------------
Message: 1
Date: Thu, 24 Mar 2016 22:06:03 +0100
From: Julien ?LIE <[email protected]>
To: [email protected]
Subject: Re: nnrpd and TCP nagle
Message-ID: <[email protected]>
Content-Type: text/plain; charset=windows-1252
Hi all,
>> I've been debugging a slow reader issue with a friend and we've found
>> that when his tin does a "listgroup", each and every article number is
>> going over the wire in its own packet. That, of course, is pretty
>> inefficient.
>
>> I've found the following in nnrpd.c (line 614):
>
>> /* Setting TCP_NODELAY to nnrpd fixes a problem of slow downloading
>> * of overviews and slow answers on some architectures (like BSD/OS). */
>> setsockopt(STDIN_FILENO, IPPROTO_TCP, TCP_NODELAY, &nodelay,
>> sizeof(nodelay));
>
>> Which, when removed, makes for way better performance, about an order
>> of magnitude down in number of packets and time for the group we've
>> been testing with.
>
>> The last time I ran INN on BSD/OS was 20 years ago, and BSD/OS has
>> been discontinued in 2003, which indicates this line is somewhat
>> historic. And for NNTP TCP_NODELAY makes no sense in the first place.
>
>> So I suggest removing this.
>
> Yup, I agree. It's almost never a good idea to tweak stuff like that with
> a modern networking stack. The kernel is probably going to be smarter
> about aggregation than we're going to be.
Couldn't we just keep it for BSD-like systems?
It was a suggestion in the thread that led to that addition in nnrpd:
http://permalink.gmane.org/gmane.network.inn/8642
The Doctor is notably still using a BSD-like system and reported that the
patch made INN way faster on his system.
Suggestion of patch:
Index: configure.ac
===================================================================
--- configure.ac (r?vision 9984)
+++ configure.ac (copie de travail)
@@ -203,6 +203,12 @@
[Define if compiling on Linux to get prototypes for some functions.])
;;
+dnl Detect BSD-based systems for later use in nnrpd code.
+*BSD*)
+ AC_DEFINE([INN_BSD], [1],
+ [Define if compiling on BSD-based systems.])
+ ;;
+
dnl HP-UX's native compiler needs a special flag to turn on ANSI, and needs
dnl -g on link as well as compile for debugging to work.
*hpux*)
Index: nnrpd/nnrpd.c
===================================================================
--- nnrpd/nnrpd.c (r?vision 9984)
+++ nnrpd/nnrpd.c (copie de travail)
@@ -13,7 +13,9 @@
#include "portable/socket.h"
#include <netdb.h>
#include <signal.h>
-#include <netinet/tcp.h>
+#if defined(INN_BSD)
+# include <netinet/tcp.h>
+#endif
#if HAVE_GETSPNAM
# include <shadow.h>
@@ -579,7 +581,6 @@
struct sockaddr *sas = (struct sockaddr *) &sss;
socklen_t length;
size_t size;
- int nodelay = 1;
memset(&Client, 0, sizeof(Client));
strlcpy(Client.host, "?", sizeof(Client.host));
@@ -648,9 +649,12 @@
Client.serverport = network_sockaddr_port(sas);
}
+#if defined(INN_BSD)
/* Setting TCP_NODELAY to nnrpd fixes a problem of slow downloading
* of overviews and slow answers on some architectures (like BSD/OS). */
+ int nodelay = 1;
setsockopt(STDIN_FILENO, IPPROTO_TCP, TCP_NODELAY, &nodelay,
sizeof(nodelay));
+#endif
notice("%s (%s) connect - port %u", Client.host, Client.ip, port);
--
Julien ?LIE
? ? Depuis quelque temps, Z?roz?rosix n'attire plus les mouches !
? Esp?rons qu'il ne nous attirera pas d'ennuis ! ? (Ast?rix)
------------------------------
Message: 2
Date: Thu, 24 Mar 2016 15:33:39 -0600 (MDT)
From: "The Doctor, 3328-138 Ave Edmonton AB T5Y 1M4, 669-2000,
473-4587" <[email protected]>
To: Julien ?LIE <[email protected]>
Cc: [email protected]
Subject: Re: nnrpd and TCP nagle
Message-ID: <[email protected]>
Content-Type: text/plain; charset="US-ASCII"
[ Charset windows-1252 unsupported, converting... ]
> Hi all,
>
> >> I've been debugging a slow reader issue with a friend and we've found
> >> that when his tin does a "listgroup", each and every article number is
> >> going over the wire in its own packet. That, of course, is pretty
> >> inefficient.
> >
> >> I've found the following in nnrpd.c (line 614):
> >
> >> /* Setting TCP_NODELAY to nnrpd fixes a problem of slow downloading
> >> * of overviews and slow answers on some architectures (like BSD/OS).
> >> */
> >> setsockopt(STDIN_FILENO, IPPROTO_TCP, TCP_NODELAY, &nodelay,
> >> sizeof(nodelay));
> >
> >> Which, when removed, makes for way better performance, about an order
> >> of magnitude down in number of packets and time for the group we've
> >> been testing with.
> >
> >> The last time I ran INN on BSD/OS was 20 years ago, and BSD/OS has
> >> been discontinued in 2003, which indicates this line is somewhat
> >> historic. And for NNTP TCP_NODELAY makes no sense in the first place.
> >
> >> So I suggest removing this.
> >
> > Yup, I agree. It's almost never a good idea to tweak stuff like that with
> > a modern networking stack. The kernel is probably going to be smarter
> > about aggregation than we're going to be.
>
> Couldn't we just keep it for BSD-like systems?
> It was a suggestion in the thread that led to that addition in nnrpd:
> http://permalink.gmane.org/gmane.network.inn/8642
>
> The Doctor is notably still using a BSD-like system and reported that the
> patch made INN way faster on his system.
>
>
> Suggestion of patch:
>
>
> Index: configure.ac
> ===================================================================
> --- configure.ac (r?vision 9984)
> +++ configure.ac (copie de travail)
> @@ -203,6 +203,12 @@
> [Define if compiling on Linux to get prototypes for some functions.])
> ;;
>
> +dnl Detect BSD-based systems for later use in nnrpd code.
> +*BSD*)
> + AC_DEFINE([INN_BSD], [1],
> + [Define if compiling on BSD-based systems.])
> + ;;
> +
> dnl HP-UX's native compiler needs a special flag to turn on ANSI, and needs
> dnl -g on link as well as compile for debugging to work.
> *hpux*)
> Index: nnrpd/nnrpd.c
> ===================================================================
> --- nnrpd/nnrpd.c (r?vision 9984)
> +++ nnrpd/nnrpd.c (copie de travail)
> @@ -13,7 +13,9 @@
> #include "portable/socket.h"
> #include <netdb.h>
> #include <signal.h>
> -#include <netinet/tcp.h>
> +#if defined(INN_BSD)
> +# include <netinet/tcp.h>
> +#endif
>
> #if HAVE_GETSPNAM
> # include <shadow.h>
> @@ -579,7 +581,6 @@
> struct sockaddr *sas = (struct sockaddr *) &sss;
> socklen_t length;
> size_t size;
> - int nodelay = 1;
>
> memset(&Client, 0, sizeof(Client));
> strlcpy(Client.host, "?", sizeof(Client.host));
> @@ -648,9 +649,12 @@
> Client.serverport = network_sockaddr_port(sas);
> }
>
> +#if defined(INN_BSD)
> /* Setting TCP_NODELAY to nnrpd fixes a problem of slow downloading
> * of overviews and slow answers on some architectures (like BSD/OS). */
> + int nodelay = 1;
> setsockopt(STDIN_FILENO, IPPROTO_TCP, TCP_NODELAY, &nodelay,
> sizeof(nodelay));
> +#endif
>
> notice("%s (%s) connect - port %u", Client.host, Client.ip, port);
>
>
>
> --
> Julien ?LIE
>
That could work.
Also can you fix the configure file for the openssl 1.1.X issue?
> ? ? Depuis quelque temps, Z?roz?rosix n'attire plus les mouches !
> ? Esp?rons qu'il ne nous attirera pas d'ennuis ! ? (Ast?rix)
> _______________________________________________
> inn-workers mailing list
> [email protected]
> https://lists.isc.org/mailman/listinfo/inn-workers
------------------------------
_______________________________________________
inn-workers mailing list
[email protected]
https://lists.isc.org/mailman/listinfo/inn-workers
End of inn-workers Digest, Vol 82, Issue 5
******************************************
