cisco 1900.,2900,3550,3560 and 3570 series switches and switch fabric

Answered Question
Dec 12th, 2008

Hi every body!

DO cisco 1900, 2900,3550,3560 and 3570 ,series switches use switch fabric like 6500 and 4500 series switches?

thanks a lot!

I have this problem too.
0 votes
Correct Answer by Joseph W. Doherty about 8 years 1 month ago

Not sure what the ports are on either the 1900 or 2800; end-of-sale models?

Let's use something that's current, such as the 2960 series. Beginning with the 2960-8TC and 2960G-8TC, the former has 8 FastEthernet ports and 1 gig port, the latter has 8 gig ports. The former spec's are 2.7 Mpps (millon packets per second) and 16 Gbps fabric, the latter spec's are 11.9 Mpps and 32 Gbps fabric.

To compute the ideal fabric bandwidth, take each ports' speed and, if duplex, double it. For the 2960-8TC that's (8 ports * 100 Mbps + 1 port * 1 Gbps) * 2 = 3.6 Gbps. For the 2960G-8TC that's 8 ports * 1 Gbps * 2 = 16 Gbps. Both these switches provide more fabric bandwidth then we need. However, if you run the same calculation for the larger gig port models, like the 2960G-24TC, it would need 48 Gbps but the fabric only supports 32 Gbps. When actual port bandwidth attempts to exceed fabric/backplane bandwidth, we'll likely see packet drops.

For pps, "wire speed" or "line rate" for 64 byte packets requires 1.488095 Mpps for 1 Gbps (divide by 10 or 100 for FastEthernet and Ethernet). Although we count bandwidth against fabric for both ingress and egress, you only need to forward them once. So we use half the bandwidth calculation we used for fabric calculation. For the 2960-8TC, that would be 1.8 Gbps * 1.488095 Mpps = (about) 2.7 Mpps (which is what this switch is rated as); for the 2960G-8TC, that would be 8 Gbps * 1.488095 Mpps = (about) 11.9 Mpps (which is also what this switch is rated as). Both these switches support maximum possible peformance. If we do the same calculation for the 2960G-48TC, it would be 48 * 1.488095 Mpps = (about) 71.4 Mpps, but since the spec notes Mpps is 39 Mpps, it could drop packets because the hardware isn't fast enough.

BTW: The highest pps rate is required for the smallest packets, so when it's quoted, you need to check for what packet sizes.

You can also run these numbers "backwards". Given 32 Gbps fabric, that will support 16 gig duplex ports at full speed. Given 39 Mpps, that would support (about) 26 Gbps full duplex if packet size were 64 bytes.

Correct Answer by Giuseppe Larosa about 8 years 1 month ago

Hello Sarah,

the numbers are half-duplex : 550,000 pps would mean as a while including rx and tx.

the same happens for the figures for fabric: so 1Gbps as to be seen able to send and receive 500 Mbps.

This is used indeed by all vendors and none will divide by two its declared performance.

In addition to this, the fabric can have an orientation: that is more performant in the direction user ports to uplinks in comparison to sending 500 Mbps between user ports.

Hope to help

Giuseppe

Correct Answer by Joseph W. Doherty about 8 years 1 month ago

Almost any LAN switch is usually described as having a "fabric", how similar it is between different platforms, would depend on the actual design, and this infomation often isn't fully documented by the vendor of the equipment. Without such documentation, it's difficult to impossible to determine how much one switch's "fabric" is like another's.

I've attached a Cisco "Switching Performance" document which shows the "Switch Fabric" bandwidth for many Cisco switches. What we're often interested in is the "fabric" bandwidth compared to ports. Ideally, the fabric bandwidth should be at least 2x the ports' bandwidth (assuming duplex ports). Less than the "ideal" is very common, which means that the switch can't substain every port moving traffic to/from another unique port at full bandwidth.

In the real world, rarely is a switch driven that hard, but matching fabric bandwidth (and forwarding pps rate) can be important to expected load on the device. For the same number of ports, this is one of the reasons we might chose different switch platforms for edge, distribution and core. Even if we chose the same chassis platform, we might populate the card slots much differently, again depending on usage. (e.g. user edge 6500 with sup32 using WS-X6148A-GE-TX vs. server edge 6500 with sup720 using WS-X6748-GE-TX w/DFC.)

  • 1
  • 2
  • 3
  • 4
  • 5
Overall Rating: 5 (3 ratings)
Loading.
Correct Answer
Joseph W. Doherty Sat, 12/13/2008 - 06:01

Almost any LAN switch is usually described as having a "fabric", how similar it is between different platforms, would depend on the actual design, and this infomation often isn't fully documented by the vendor of the equipment. Without such documentation, it's difficult to impossible to determine how much one switch's "fabric" is like another's.

I've attached a Cisco "Switching Performance" document which shows the "Switch Fabric" bandwidth for many Cisco switches. What we're often interested in is the "fabric" bandwidth compared to ports. Ideally, the fabric bandwidth should be at least 2x the ports' bandwidth (assuming duplex ports). Less than the "ideal" is very common, which means that the switch can't substain every port moving traffic to/from another unique port at full bandwidth.

In the real world, rarely is a switch driven that hard, but matching fabric bandwidth (and forwarding pps rate) can be important to expected load on the device. For the same number of ports, this is one of the reasons we might chose different switch platforms for edge, distribution and core. Even if we chose the same chassis platform, we might populate the card slots much differently, again depending on usage. (e.g. user edge 6500 with sup32 using WS-X6148A-GE-TX vs. server edge 6500 with sup720 using WS-X6748-GE-TX w/DFC.)

sarahr202 Sat, 12/13/2008 - 09:14

Thanks a lot Joseph for excellent explanation and for your time!

sarahr202 Sat, 12/13/2008 - 09:31

Just one question, based on your attachement.

For example for platform 1900/2800

switch performance 550,000 Mpps.

switch fabric 1 Gbps.

How do we interpret the above? Should I interpret as the 1900/2800 platforms have 1Gbps backplane bandwidth and can forward up to 550,000 Mpps( mega packet per second)?

Or this number also include both, the number of packet that can be received and forwarded by these platforms in one second?

Thanks a lot and have a nice weekend!

Correct Answer
Giuseppe Larosa Sat, 12/13/2008 - 11:17

Hello Sarah,

the numbers are half-duplex : 550,000 pps would mean as a while including rx and tx.

the same happens for the figures for fabric: so 1Gbps as to be seen able to send and receive 500 Mbps.

This is used indeed by all vendors and none will divide by two its declared performance.

In addition to this, the fabric can have an orientation: that is more performant in the direction user ports to uplinks in comparison to sending 500 Mbps between user ports.

Hope to help

Giuseppe

sarahr202 Sun, 12/14/2008 - 14:32

thanks a lot Giuseppe!

I read cisco manuals which describe pps as the number of packets can be forwarded per second by a switch.That means we are forcussing only in one direction which is transmission not reception of packets.

Based on your reply, I should interpret 550,000 pps as the number of packets that can be transmitted or received by a switch in a second.

Based on Cisco Manual, 550,000 pps means upto 550,000 packets can be forwarded in one second by switch.

I would really appreciate if you elaborate on this.

thanks a lot!

Below is the link, please look at end of the page where performance is described.

http://www.buy.com/prod/cisco-catalyst-2960g-24tc-ethernet-switch-20-x-10-100-1000base-t-lan-4/q/loc/101/202074213.html

Thanks a lot!

Correct Answer
Joseph W. Doherty Sat, 12/13/2008 - 16:44

Not sure what the ports are on either the 1900 or 2800; end-of-sale models?

Let's use something that's current, such as the 2960 series. Beginning with the 2960-8TC and 2960G-8TC, the former has 8 FastEthernet ports and 1 gig port, the latter has 8 gig ports. The former spec's are 2.7 Mpps (millon packets per second) and 16 Gbps fabric, the latter spec's are 11.9 Mpps and 32 Gbps fabric.

To compute the ideal fabric bandwidth, take each ports' speed and, if duplex, double it. For the 2960-8TC that's (8 ports * 100 Mbps + 1 port * 1 Gbps) * 2 = 3.6 Gbps. For the 2960G-8TC that's 8 ports * 1 Gbps * 2 = 16 Gbps. Both these switches provide more fabric bandwidth then we need. However, if you run the same calculation for the larger gig port models, like the 2960G-24TC, it would need 48 Gbps but the fabric only supports 32 Gbps. When actual port bandwidth attempts to exceed fabric/backplane bandwidth, we'll likely see packet drops.

For pps, "wire speed" or "line rate" for 64 byte packets requires 1.488095 Mpps for 1 Gbps (divide by 10 or 100 for FastEthernet and Ethernet). Although we count bandwidth against fabric for both ingress and egress, you only need to forward them once. So we use half the bandwidth calculation we used for fabric calculation. For the 2960-8TC, that would be 1.8 Gbps * 1.488095 Mpps = (about) 2.7 Mpps (which is what this switch is rated as); for the 2960G-8TC, that would be 8 Gbps * 1.488095 Mpps = (about) 11.9 Mpps (which is also what this switch is rated as). Both these switches support maximum possible peformance. If we do the same calculation for the 2960G-48TC, it would be 48 * 1.488095 Mpps = (about) 71.4 Mpps, but since the spec notes Mpps is 39 Mpps, it could drop packets because the hardware isn't fast enough.

BTW: The highest pps rate is required for the smallest packets, so when it's quoted, you need to check for what packet sizes.

You can also run these numbers "backwards". Given 32 Gbps fabric, that will support 16 gig duplex ports at full speed. Given 39 Mpps, that would support (about) 26 Gbps full duplex if packet size were 64 bytes.

Jonn cos Thu, 10/17/2013 - 08:16

Dear Sir,

Please tell me from where this value is derived ?

1.488095 Mpps

Actions

This Discussion