On Wednesday, July 09, 1997 9:34 PM, Josh Beck [SMTP:jbeck@connectnet.com] wrote:
Hello, I just thought of something. We are in the process of purchasing a 4 Mb CIR from another backbone. Now, we have the choice of ATM or standard T3 delivery (over a DS3 either way). Now, if we get ATM, that 4 Mb CIR turns into:
[ (53-5)/53 ] * 4 Mb/s = 48/53 * 4 Mb/s = 3.62 Mb/s
Emperical data shows that we are currently losing about 20.5% of capacity to IP over ATM overhead on fairly aggregated traffic. This means that *IF* your new connection is being measured as 4Mbps of cell bandwisth, you will only be getting 3.18Mbps. You may want to verify from the company providing this link what exactly are they limiting you to? btw, the extra overhead is lost in things like the last cell of a packet not being full, etc. Chris A. Icide Sr. Engineer Nap.Net, L.L.C.
On Tue, Jul 09, 1996 at 10:38:57PM -0500, Chris A. Icide wrote:
On Wednesday, July 09, 1997 9:34 PM, Josh Beck [SMTP:jbeck@connectnet.com] wrote:
Hello, I just thought of something. We are in the process of purchasing a 4 Mb CIR from another backbone. Now, we have the choice of ATM or standard T3 delivery (over a DS3 either way). Now, if we get ATM, that 4 Mb CIR turns into:
[ (53-5)/53 ] * 4 Mb/s = 48/53 * 4 Mb/s = 3.62 Mb/s
Emperical data shows that we are currently losing about 20.5% of capacity to IP over ATM overhead on fairly aggregated traffic. This means that *IF* your new connection is being measured as 4Mbps of cell bandwisth, you will only be getting 3.18Mbps. You may want to verify from the company providing this link what exactly are they limiting you to?
btw, the extra overhead is lost in things like the last cell of a packet not being full, etc.
Chris A. Icide Sr. Engineer Nap.Net, L.L.C.
My God, someone admits it? I've used 20% as the general ATM overhead now for almost two years, and have been poo-pooed by lots of people claiming that it wasn't anywhere near that bad. Funny how it all comes out in the end. :-) -- -- Karl Denninger (karl@MCS.Net)| MCSNet - The Finest Internet Connectivity http://www.mcs.net/~karl | T1's from $600 monthly to FULL DS-3 Service | 99 Analog numbers, 77 ISDN, http://www.mcs.net/ Voice: [+1 312 803-MCS1 x219]| NOW Serving 56kbps DIGITAL on our analog lines! Fax: [+1 312 803-4929] | 2 FULL DS-3 Internet links; 400Mbps B/W Internal
i've never heard anything *less* than 20% loss in ATM overhead. On Wed, 9 Jul 1997, Karl Denninger wrote:
On Tue, Jul 09, 1996 at 10:38:57PM -0500, Chris A. Icide wrote:
On Wednesday, July 09, 1997 9:34 PM, Josh Beck [SMTP:jbeck@connectnet.com] wrote:
Hello, I just thought of something. We are in the process of purchasing a 4 Mb CIR from another backbone. Now, we have the choice of ATM or standard T3 delivery (over a DS3 either way). Now, if we get ATM, that 4 Mb CIR turns into:
[ (53-5)/53 ] * 4 Mb/s = 48/53 * 4 Mb/s = 3.62 Mb/s
Emperical data shows that we are currently losing about 20.5% of capacity to IP over ATM overhead on fairly aggregated traffic. This means that *IF* your new connection is being measured as 4Mbps of cell bandwisth, you will only be getting 3.18Mbps. You may want to verify from the company providing this link what exactly are they limiting you to?
btw, the extra overhead is lost in things like the last cell of a packet not being full, etc.
Chris A. Icide Sr. Engineer Nap.Net, L.L.C.
My God, someone admits it?
I've used 20% as the general ATM overhead now for almost two years, and have been poo-pooed by lots of people claiming that it wasn't anywhere near that bad.
Funny how it all comes out in the end. :-)
-- -- Karl Denninger (karl@MCS.Net)| MCSNet - The Finest Internet Connectivity http://www.mcs.net/~karl | T1's from $600 monthly to FULL DS-3 Service | 99 Analog numbers, 77 ISDN, http://www.mcs.net/ Voice: [+1 312 803-MCS1 x219]| NOW Serving 56kbps DIGITAL on our analog lines! Fax: [+1 312 803-4929] | 2 FULL DS-3 Internet links; 400Mbps B/W Internal
Question: On a Cisco with an AIP card, how do you determain what the overhead is? We connect to our upstream provider via ATM. .stb On Thu, 10 Jul 1997, Ben Black wrote:
i've never heard anything *less* than 20% loss in ATM overhead.
On Wed, 9 Jul 1997, Karl Denninger wrote:
On Tue, Jul 09, 1996 at 10:38:57PM -0500, Chris A. Icide wrote:
On Wednesday, July 09, 1997 9:34 PM, Josh Beck [SMTP:jbeck@connectnet.com] wrote:
Hello, I just thought of something. We are in the process of purchasing a 4 Mb CIR from another backbone. Now, we have the choice of ATM or standard T3 delivery (over a DS3 either way). Now, if we get ATM, that 4 Mb CIR turns into:
[ (53-5)/53 ] * 4 Mb/s = 48/53 * 4 Mb/s = 3.62 Mb/s
Emperical data shows that we are currently losing about 20.5% of capacity to IP over ATM overhead on fairly aggregated traffic. This means that *IF* your new connection is being measured as 4Mbps of cell bandwisth, you will only be getting 3.18Mbps. You may want to verify from the company providing this link what exactly are they limiting you to?
btw, the extra overhead is lost in things like the last cell of a packet not being full, etc.
Chris A. Icide Sr. Engineer Nap.Net, L.L.C.
My God, someone admits it?
I've used 20% as the general ATM overhead now for almost two years, and have been poo-pooed by lots of people claiming that it wasn't anywhere near that bad.
Funny how it all comes out in the end. :-)
-- -- Karl Denninger (karl@MCS.Net)| MCSNet - The Finest Internet Connectivity http://www.mcs.net/~karl | T1's from $600 monthly to FULL DS-3 Service | 99 Analog numbers, 77 ISDN, http://www.mcs.net/ Voice: [+1 312 803-MCS1 x219]| NOW Serving 56kbps DIGITAL on our analog lines! Fax: [+1 312 803-4929] | 2 FULL DS-3 Internet links; 400Mbps B/W Internal
At 04:06 PM 7/10/97 -0400, Stephen Balbach wrote:
Question: On a Cisco with an AIP card, how do you determain what the overhead is? We connect to our upstream provider via ATM.
Again, from experience, a DS3 AIP is "full" when the 5-minute-average counters are reporting ~34Mpbs (in other words, the counters never went higher). Performance wasn't terrific, but if you're looking for raw quantity to compute "efficiency", or what some call goodput, then my crayon on painted wall calculation is ~34Mbps (observed max) / ~45Mpbs (approx DS3 raw max) = ~75% "efficiency". Given that networks and data have a measurable coefficient of friction, I'd say that's the max you could get under optimal conditions is something less. It's not an option with the AIPs, but when we ran switch to switch we could get another 4 - 5 Mpbs out of a DS3 by turning off PLCP (not an endorsement, recommendation, or even technically sound, but it worked). Another way of looking at this is that we know an ATM PVC over a DS3 using PLCP is configurable for a maximum of 96000 cells per second. 96000 cells per second * 48 payload octets per cell * 8 bits per octet = 36864000 bps, or 36.864 Mbps, not too far from what I observed above. With PLCP turned off, the maximum PVC config was ~105000 cells per second, yielding ~40.3 Mpbs. So AIP to AIP has the downside of all the protocol overhead with no benefit in a point to point connection. But you're trading off for the convenience of not having to use a HSSI port to a T3 CSU/DSU with the additional rack space and power concerns. And if I had to pick one particular device which I spent the most time fussing with and repairing/replacing, it would be T3 CSU/DSUs. -peter
.stb
On Thu, 10 Jul 1997, Ben Black wrote:
i've never heard anything *less* than 20% loss in ATM overhead.
On Wed, 9 Jul 1997, Karl Denninger wrote:
On Tue, Jul 09, 1996 at 10:38:57PM -0500, Chris A. Icide wrote:
On Wednesday, July 09, 1997 9:34 PM, Josh Beck
wrote:
Hello, I just thought of something. We are in the process of purchasing a 4 Mb CIR from another backbone. Now, we have the choice of ATM or standard T3 delivery (over a DS3 either way). Now, if we get ATM, that 4 Mb CIR turns into:
[ (53-5)/53 ] * 4 Mb/s = 48/53 * 4 Mb/s = 3.62 Mb/s
Emperical data shows that we are currently losing about 20.5% of capacity to IP over ATM overhead on fairly aggregated traffic. This means
your new connection is being measured as 4Mbps of cell bandwisth, you will only be getting 3.18Mbps. You may want to verify from the company providing this link what exactly are they limiting you to?
btw, the extra overhead is lost in things like the last cell of a
being full, etc.
Chris A. Icide Sr. Engineer Nap.Net, L.L.C.
My God, someone admits it?
I've used 20% as the general ATM overhead now for almost two years, and have been poo-pooed by lots of people claiming that it wasn't anywhere near
[SMTP:jbeck@connectnet.com] that *IF* packet not that
bad.
Funny how it all comes out in the end. :-)
-- -- Karl Denninger (karl@MCS.Net)| MCSNet - The Finest Internet Connectivity http://www.mcs.net/~karl | T1's from $600 monthly to FULL DS-3 Service | 99 Analog numbers, 77 ISDN, http://www.mcs.net/ Voice: [+1 312 803-MCS1 x219]| NOW Serving 56kbps DIGITAL on our analog lines! Fax: [+1 312 803-4929] | 2 FULL DS-3 Internet links; 400Mbps B/W Internal
PLCP mapping is the only option on a Cisco AIP (DS-3), as you probably already know.. if you have a ATM Switch that is current, you can change the mapping to Direct Cell mapping, which will give you about 3 megs more of bandwidth on the line rate.. All of our switch-switch DS-3's are direct mapping, and all of our switch-routers are OC-3 SONET, so you don't have the same loss of bandwidth as PLCP on a DS-3. (also easier to aggregate multiple backbone circuits on one card). Eric _______________________________________________________ Eric D. Madison - Senior Network Engineer - ACSI - Advanced Data Services - ATM/IP Backbone Group 24 Hour NMC/NOC (800)291-7889 Email: noc@acsi.net On Thu, 10 Jul 1997, Peter Kline wrote:
At 04:06 PM 7/10/97 -0400, Stephen Balbach wrote:
Question: On a Cisco with an AIP card, how do you determain what the overhead is? We connect to our upstream provider via ATM.
Again, from experience, a DS3 AIP is "full" when the 5-minute-average counters are reporting ~34Mpbs (in other words, the counters never went higher). Performance wasn't terrific, but if you're looking for raw quantity to compute "efficiency", or what some call goodput, then my crayon on painted wall calculation is ~34Mbps (observed max) / ~45Mpbs (approx DS3 raw max) = ~75% "efficiency". Given that networks and data have a measurable coefficient of friction, I'd say that's the max you could get under optimal conditions is something less.
It's not an option with the AIPs, but when we ran switch to switch we could get another 4 - 5 Mpbs out of a DS3 by turning off PLCP (not an endorsement, recommendation, or even technically sound, but it worked).
Another way of looking at this is that we know an ATM PVC over a DS3 using PLCP is configurable for a maximum of 96000 cells per second. 96000 cells per second * 48 payload octets per cell * 8 bits per octet = 36864000 bps, or 36.864 Mbps, not too far from what I observed above. With PLCP turned off, the maximum PVC config was ~105000 cells per second, yielding ~40.3 Mpbs.
So AIP to AIP has the downside of all the protocol overhead with no benefit in a point to point connection. But you're trading off for the convenience of not having to use a HSSI port to a T3 CSU/DSU with the additional rack space and power concerns. And if I had to pick one particular device which I spent the most time fussing with and repairing/replacing, it would be T3 CSU/DSUs.
-peter
.stb
On Thu, 10 Jul 1997, Ben Black wrote:
i've never heard anything *less* than 20% loss in ATM overhead.
On Wed, 9 Jul 1997, Karl Denninger wrote:
On Tue, Jul 09, 1996 at 10:38:57PM -0500, Chris A. Icide wrote:
On Wednesday, July 09, 1997 9:34 PM, Josh Beck
wrote:
Hello, I just thought of something. We are in the process of purchasing a 4 Mb CIR from another backbone. Now, we have the choice of ATM or standard T3 delivery (over a DS3 either way). Now, if we get ATM, that 4 Mb CIR turns into:
[ (53-5)/53 ] * 4 Mb/s = 48/53 * 4 Mb/s = 3.62 Mb/s
Emperical data shows that we are currently losing about 20.5% of capacity to IP over ATM overhead on fairly aggregated traffic. This means
your new connection is being measured as 4Mbps of cell bandwisth, you will only be getting 3.18Mbps. You may want to verify from the company providing this link what exactly are they limiting you to?
btw, the extra overhead is lost in things like the last cell of a
being full, etc.
Chris A. Icide Sr. Engineer Nap.Net, L.L.C.
My God, someone admits it?
I've used 20% as the general ATM overhead now for almost two years, and have been poo-pooed by lots of people claiming that it wasn't anywhere near
[SMTP:jbeck@connectnet.com] that *IF* packet not that
bad.
Funny how it all comes out in the end. :-)
-- -- Karl Denninger (karl@MCS.Net)| MCSNet - The Finest Internet Connectivity http://www.mcs.net/~karl | T1's from $600 monthly to FULL DS-3 Service | 99 Analog numbers, 77 ISDN, http://www.mcs.net/ Voice: [+1 312 803-MCS1 x219]| NOW Serving 56kbps DIGITAL on our analog lines! Fax: [+1 312 803-4929] | 2 FULL DS-3 Internet links; 400Mbps B/W Internal
I've used 20% as the general ATM overhead now for almost two years, and have been poo-pooed by lots of people claiming that it wasn't anywhere near that bad.
Our inhouse figure is 23%. That was an empirical test of a traffic generator comparing performance to Cicso HDLC. This might be slightly pessimistic as Cisco traffic shaping (yes, even though you can't see it, it is there working) is really very aggressive and with bursty traffic it tends to prefer underfilling the line to overfilling it. Alex Bligh Xara Networks
On Fri, 11 Jul 1997, Alex.Bligh wrote:
I've used 20% as the general ATM overhead now for almost two years, and have been poo-pooed by lots of people claiming that it wasn't anywhere near that bad.
Our inhouse figure is 23%. That was an empirical test of a traffic generator comparing performance to Cicso HDLC. This might be slightly pessimistic as Cisco traffic shaping (yes, even though you can't see it, it is there working) is really very aggressive and with bursty traffic it tends to prefer underfilling the line to overfilling it.
I'm not sure if this is pessimistic. When we ran calculations based on packet size distribution of CICNet, we got 23% as well, and I believe Peter's calculations based on ICM traffic agreed with this also. -dorian
participants (8)
-
Alex.Bligh
-
Ben Black
-
Chris A. Icide
-
Dorian R. Kim
-
Eric D. Madison
-
Karl Denninger
-
Peter Kline
-
Stephen Balbach