Proposed patch to iproute to add Initial Max Congestion Window Size route tuning...

Vince Worthington wrote:
> Hello,
>
> We've been working with a customer of ours who was experiencing some
> latency issues, and in the process of helping them solve their problem
> we found that adding the ability to adjust the Initial Max Congestion
> Window size on a route, they were able to tune their routes to perform
> more favorably in their mixed Linux and Solaris environment.  There
> were a couple of other tuning steps necessary as well but we did find
> that the ability to set a larger initial max congestion window size
> was helpful in making Linux behave more like Solaris in low-latency
> situations.
>
> The kernel already supports this route attribute, this patch simply
> adds the option to iproute.c to provide a means to set it from userspace.
>
> I've attached the patch we used in test builds of the iproute packages
> we distribute in RHEL3 and RHEL4 for your consideration for possible
> acceptance and inclusion in iproute.  Please let us know if there is
> any further information you would need or if there is a more
> appropriate venue to submit this patch.
>
> Thanks,
> Vince Worthington
> Red Hat, Inc.

Stephen,

I apologize for the last patch I sent not applying cleanly to the latest
iproute2 sources.  I realized that this might impede any
review/acceptance of the patch and created a separate patch against the
20060323 build of iproute2, which seems to be the latest (by looking at
the website).

The previous patch wouldn't apply cleanly due to the label of the
ssthresh argument being corrected from REALMS to NUMBER.

Please let us know if there is any other information or assistance we
can be with this patch submission.

Thanks
Vince

diff --git a/ip/iproute.c b/ip/iproute.c
index a43c09e6cba6a82de0a45e527780e19ff2e9befa..0b70347325354fbc24480fdcf289add1c70c2c41 100644 (file)
--- a/ip/iproute.c
+++ b/ip/iproute.c
@@ -59,8 +59,8 @@ static void usage(void)
        fprintf(stderr, "NH := [ via ADDRESS ] [ dev STRING ] [ weight NUMBER ] NHFLAGS\n");
        fprintf(stderr, "OPTIONS := FLAGS [ mtu NUMBER ] [ advmss NUMBER ]\n");
        fprintf(stderr, "           [ rtt NUMBER ] [ rttvar NUMBER ]\n");
-       fprintf(stderr, "           [ window NUMBER] [ cwnd NUMBER ] [ ssthresh NUMBER ]\n");
-       fprintf(stderr, "           [ realms REALM ]\n");
+       fprintf(stderr, "           [ window NUMBER] [ cwnd NUMBER ] [ initcwnd NUMBER ]\n");
+       fprintf(stderr, "           [ ssthresh NUMBER ] [ realms REALM ]\n");
        fprintf(stderr, "TYPE := [ unicast | local | broadcast | multicast | throw |\n");
        fprintf(stderr, "          unreachable | prohibit | blackhole | nat ]\n");
        fprintf(stderr, "TABLE_ID := [ local | main | default | all | NUMBER ]\n");
@@ -487,6 +487,15 @@ int print_route(const struct sockaddr_nl *who, struct nlmsghdr *n, void *arg)
                        mxlock = *(unsigned*)RTA_DATA(mxrta[RTAX_LOCK]);
 
                for (i=2; i<=RTAX_MAX; i++) {
+
+                       /*
+                        * "hoplimit" and "features" are not supported in
+                        * userspace yet, but they are present in kernel's
+                        * RTM_METRICS array in include/linux/rtnetlink.h,
+                        * so putting placeholders in here for now so we
+                        * can match the position of initcwnd in this
+                        * structure:                                   */
+
                        static char *mx_names[] = 
                        {
                                "mtu",
@@ -497,6 +506,9 @@ int print_route(const struct sockaddr_nl *who, struct nlmsghdr *n, void *arg)
                                "cwnd",
                                "advmss",
                                "reordering",
+                               "hoplimit",
+                               "initcwnd",
+                               "features",
                        };
                        static int hz;
                        if (mxrta[i] == NULL)
@@ -794,6 +806,16 @@ int iproute_modify(int cmd, unsigned flags, int argc, char **argv)
                        if (get_unsigned(&win, *argv, 0))
                                invarg("\"cwnd\" value is invalid\n", *argv);
                        rta_addattr32(mxrta, sizeof(mxbuf), RTAX_CWND, win);
+               } else if (matches(*argv, "initcwnd") == 0) {
+                       unsigned win;
+                       NEXT_ARG();
+                       if (strcmp(*argv, "lock") == 0) {
+                               mxlock |= (1<<RTAX_INITCWND);
+                               NEXT_ARG();
+                       }
+                       if (get_unsigned(&win, *argv, 0))
+                               invarg("\"initcwnd\" value is invalid\n", *argv);
+                       rta_addattr32(mxrta, sizeof(mxbuf), RTAX_INITCWND, win);
                } else if (matches(*argv, "rttvar") == 0) {
                        unsigned win;
                        NEXT_ARG();