asterisk users - Jul 2004 - Multiple E1s over TDMoE?

When I was trying to run mutiple E1s over TDMoE, the zaptel would
drivers complain about too little memory, whenever I added more than 31
channels. Requesting 62 channels in a dynamic span gave me

... span creation failed: Cannot allocate memory

upon loading the zaptel drivers.

How would you go about running, 8 or 16 say, E1s over TDMoE? Would you
create multiple dynamic spans or just one large one? How would you
assign d channels to spans, if you had just one large span?

Did any of you guys try this before?

Thilo

Thilo Salmon wrote:
> How would you go about running, 8 or 16 say, E1s over TDMoE? Would you
> create multiple dynamic spans or just one large one? How would you
> assign d channels to spans, if you had just one large span?
I would be surprised if the drivers supported more than 31 channels in a 
single span; these are just virtual PRI spans after all, so there's no 
reason for the driver to support an unlimited number of channels per span.

On Sat, Jul 03, 2004 at 07:44:54PM +0200, Thilo Salmon
wrote:
>
> How would you go about running, 8 or 16 say, E1s over TDMoE? Would you
> create multiple dynamic spans or just one large one? How would you
> assign d channels to spans, if you had just one large span?
> 
> Did any of you guys try this before?
> 
Somewhat, yes.

I run 4 TDMoE E1s between pairs of servers (4 E1s between each of
them).

What I think about that :
- it works
- TDMoE doesn't like SMP. It doesn't like running on a NIC used for
  other kind of traffic. It will crash your box under heavy (non
  TDMoE) load. I believe that there must be some race condition
  related to dev_queue_xmit(), which is probably not callable at
  anytime.
- the subaddr support is not complete in the released driver. Here is
  a patch that will handle it (as described in zaptel.conf)
- it is not 100% reliable. You will get frame drops, and you will
  notice it if you look at your D-channel dumps.
- use high quality NICs and switches.

-- 
Nicolas Bougues
Axialys Interactive
-------------- next part --------------
--- ztd-eth.c.old	2004-02-01 06:53:58.000000000 +0100
+++ ztd-eth.c	2004-07-11 00:51:45.000000000 +0200
@@ -251,7 +251,7 @@
 {
 	struct ztdeth *z;
 	char src[256];
-	char tmp[256], *tmp2, *tmp3;
+	char tmp[256], *tmp2, *tmp3, *tmp4 = NULL;
 	int res,x;
 	unsigned long flags;
 
@@ -273,6 +273,7 @@
 			return NULL;
 		}
 		if (tmp2) {
+			tmp4 = strchr(tmp2+1, '/') +1;
 			/* We don't have SSCANF :(  Gotta do this the hard way */
 			tmp3 = strchr(tmp2, ':');
 			for (x=0;x<6;x++) {
@@ -288,7 +289,8 @@
 				} else
 					break;
 				if ((tmp2 = tmp3))
-					tmp3 = strchr(tmp2, ':');
+					if (!(tmp3 = strchr (tmp2, ':')))
+						tmp3 = strchr (tmp2, '/');
 			}
 			if (x != 6) {
 				printk("TDMoE: Invalid MAC address in: %s\n", addr);
@@ -300,6 +302,25 @@
 			kfree(z);
 			return NULL;
 		}
+		if (tmp4) {
+			int sub = 0;
+			int mul = 1;
+
+			// We have a subaddr
+			tmp3 = tmp4 + strlen (tmp4) - 1;
+			while (tmp3 >= tmp4) {
+				if (*tmp3 >= '0' && *tmp3 <= '9') {
+					sub += (*tmp3 - '0') * mul;
+				} else {
+					printk("TDMoE: Invalid subaddress\n");
+					kfree(z);
+					return NULL;
+				}
+				mul *= 10;
+				tmp3--;
+			}
+			z->subaddr = htons(sub);
+		}
 		z->dev = dev_get_by_name(z->ethdev);
 		if (!z->dev) {
 			printk("TDMoE: Invalid device '%s'\n", z->ethdev);
@@ -311,7 +332,7 @@
 		for (x=0;x<5;x++)
 			sprintf(src + strlen(src), "%02x:", z->dev->dev_addr[x]);
 		sprintf(src + strlen(src), "%02x", z->dev->dev_addr[5]);
-		printk("TDMoE: Added new interface for %s at %s (addr=%s,
src=%s)\n", span->name, z->dev->name, addr, src);
+		printk("TDMoE: Added new interface for %s at %s (addr=%s, src=%s,
subaddr(net byte order)=%d)\n", span->name, z->dev->name, addr,
src, z->subaddr);
 			
 		spin_lock_irqsave(&zlock, flags);
 		z->next = zdevs;
@@ -350,3 +371,6 @@
 
 module_init(ztdeth_init);
 module_exit(ztdeth_exit);