Header Only - DO NOT REMOVE - Extreme Networks

How to display maximum throughput of the routing engine on a XOS switch


We would like to know how much data is going through the routing engine of a BD8800, in other words how much traffic is routed by the switch (the actual maximum we would like to learn)

9 replies

Userlevel 7
Hi George,
Welcome to The Hub!

If you haven't already, take a look at Pages 5 and 6 of the BlackDiamond 8000 Series Datasheet. Backplane and slot capacity are listed there, if that is what you're looking for. There's also some additional capacity information on page 11. If you are looking for something else, let us know!

-Drew
Hello Drew,

Thanks for your reply. Yes, but these are the theoretical numbers.
We would like to know the actual ammount of routed traffic on our blackdiamond, or better yet what is the maximum ammount of routed traffic our blackdiamond has to deal with, the peek moment.

George
Userlevel 7
George van Assen wrote:

Hello Drew,

Thanks for your reply. Yes, but these are the theoretical numbers.
We would like to know the actual ammount of routed traffic on our blackdiamond, or better yet what is the maximum ammount of routed traffic our blackdiamond has to deal with, the peek moment.

George

Hi George,
Can you send me the output of
code:
show slot
from one of your systems. The numbers can vary based on chassis model and type of MSM and I/O module.
From that, we'll see what we can get together for you.

Thanks!
George van Assen wrote:

Hello Drew,

Thanks for your reply. Yes, but these are the theoretical numbers.
We would like to know the actual ammount of routed traffic on our blackdiamond, or better yet what is the maximum ammount of routed traffic our blackdiamond has to deal with, the peek moment.

George

show sl

[KBD-CR1A.2 # show slot



Slots Type Configured State Ports Flags

-------------------------------------------------------------------------------

Slot-1 G48Tc G48Tc Operational 48 M

Slot-2 G48Tc G48Tc Operational 48 M

Slot-3 G48Tc G48Tc Operational 48 M

Slot-4 Empty 0

Slot-5 Empty 0

Slot-6 Empty 0

Slot-7 10G4Xc 10G4Xc Operational 4 M

Slot-8 Empty 0

Slot-9 Empty 0

Slot-10 G48Xc G48Xc Operational 48 M

MSM-A MSM-48c Operational 0

MSM-B Empty 0



Flags : M - Backplane link to Master is Active

B - Backplane link to Backup is also Active

D - Slot Disabled

I - Insufficient Power (refer to "show power budget")

BD-CR1A.3 #
Hello Drew,

Above the output of the show slot command.. Excuse me for my late reply but had a internet free day yesterday...

George
Userlevel 7
Hi George,

Routing is performed in HW and you can achieved the same throughput in L3 than in L2. Because everything happens in HW, to keep that L3 throughput, you need to stay within the HW capacity of the L3 tables, otherwise slow path (CPU) may happen.

Following the link Drew gave you, you can find all the relevant information.

Looking at the first table of the DS on page 5: You have a BD8810 with c-series modules and a single MSM-48c. You have then a backplane bandwidth capacity of 24Gbps per Slot. You can double that adding a second MSM-48c.

Does it mean the max throughput is 24 * nb_slots? Yes and no. It depends on the traffic pattern and how many L3 entries are considered.

- What about the traffic pattern? I/O module can perform local forwarding, so if your destination is on the same chip, the backplane is not used. Worst case would be every single packet has to go through the backplane.

- What about L3 tables? Simplifying a bit, I/O modules store all the forwarding information in their chip. As long as you stay within their HW capacity, traffic is HW forwarded. But if you put too many entries (too many LPM routes, too many ARP entries, Multicast...) then the chip will not be able to HW forward everything, and the CPU will be used.

You can issue that CLI command to have a view of your L3 resources usage. It's better to execute it during a normal day with a normal activity:

sh iproute reserved-entries statistics

Your limits is listed on the same table of the DS, and this output will also give you the limits. It could be useful to know what EXOS version you're using.

So, assuming you're within the HW capacity, your worst case throughput is 24*5 (slots used) = 120Gbps (or 240Gbps in Marketing math).
Grosjean, Stephane wrote:

Hi George,

Routing is performed in HW and you can achieved the same throughput in L3 than in L2. Because everything happens in HW, to keep that L3 throughput, you need to stay within the HW capacity of the L3 tables, otherwise slow path (CPU) may happen.

Following the link Drew gave you, you can find all the relevant information.

Looking at the first table of the DS on page 5: You have a BD8810 with c-series modules and a single MSM-48c. You have then a backplane bandwidth capacity of 24Gbps per Slot. You can double that adding a second MSM-48c.

Does it mean the max throughput is 24 * nb_slots? Yes and no. It depends on the traffic pattern and how many L3 entries are considered.

- What about the traffic pattern? I/O module can perform local forwarding, so if your destination is on the same chip, the backplane is not used. Worst case would be every single packet has to go through the backplane.

- What about L3 tables? Simplifying a bit, I/O modules store all the forwarding information in their chip. As long as you stay within their HW capacity, traffic is HW forwarded. But if you put too many entries (too many LPM routes, too many ARP entries, Multicast...) then the chip will not be able to HW forward everything, and the CPU will be used.

You can issue that CLI command to have a view of your L3 resources usage. It's better to execute it during a normal day with a normal activity:

sh iproute reserved-entries statistics

Your limits is listed on the same table of the DS, and this output will also give you the limits. It could be useful to know what EXOS version you're using.

So, assuming you're within the HW capacity, your worst case throughput is 24*5 (slots used) = 120Gbps (or 240Gbps in Marketing math).

Thanks for your reply. Again I am not looking for theoretical values of the capacity of this switch.
I am looking for the maximum amount of routed traffic on the switch during normal operation.
Correct me if I am wrong but the "show iproute reserved-entries statistics" is more a summary of the ammount of routes in the route table...
Userlevel 7
Grosjean, Stephane wrote:

Hi George,

Routing is performed in HW and you can achieved the same throughput in L3 than in L2. Because everything happens in HW, to keep that L3 throughput, you need to stay within the HW capacity of the L3 tables, otherwise slow path (CPU) may happen.

Following the link Drew gave you, you can find all the relevant information.

Looking at the first table of the DS on page 5: You have a BD8810 with c-series modules and a single MSM-48c. You have then a backplane bandwidth capacity of 24Gbps per Slot. You can double that adding a second MSM-48c.

Does it mean the max throughput is 24 * nb_slots? Yes and no. It depends on the traffic pattern and how many L3 entries are considered.

- What about the traffic pattern? I/O module can perform local forwarding, so if your destination is on the same chip, the backplane is not used. Worst case would be every single packet has to go through the backplane.

- What about L3 tables? Simplifying a bit, I/O modules store all the forwarding information in their chip. As long as you stay within their HW capacity, traffic is HW forwarded. But if you put too many entries (too many LPM routes, too many ARP entries, Multicast...) then the chip will not be able to HW forward everything, and the CPU will be used.

You can issue that CLI command to have a view of your L3 resources usage. It's better to execute it during a normal day with a normal activity:

sh iproute reserved-entries statistics

Your limits is listed on the same table of the DS, and this output will also give you the limits. It could be useful to know what EXOS version you're using.

So, assuming you're within the HW capacity, your worst case throughput is 24*5 (slots used) = 120Gbps (or 240Gbps in Marketing math).

Hi,

The theoretical values can be achieved. But maybe what you want to know is how much traffic is _currently_ routed on your switch?

The "sh ipr reserved stat" command will tell you how many LPM/Hosts/Multicast entries are currently programmed on every slots. This is not the routing table, but really the HW (not a software view, a hardware view).
Grosjean, Stephane wrote:

Hi George,

Routing is performed in HW and you can achieved the same throughput in L3 than in L2. Because everything happens in HW, to keep that L3 throughput, you need to stay within the HW capacity of the L3 tables, otherwise slow path (CPU) may happen.

Following the link Drew gave you, you can find all the relevant information.

Looking at the first table of the DS on page 5: You have a BD8810 with c-series modules and a single MSM-48c. You have then a backplane bandwidth capacity of 24Gbps per Slot. You can double that adding a second MSM-48c.

Does it mean the max throughput is 24 * nb_slots? Yes and no. It depends on the traffic pattern and how many L3 entries are considered.

- What about the traffic pattern? I/O module can perform local forwarding, so if your destination is on the same chip, the backplane is not used. Worst case would be every single packet has to go through the backplane.

- What about L3 tables? Simplifying a bit, I/O modules store all the forwarding information in their chip. As long as you stay within their HW capacity, traffic is HW forwarded. But if you put too many entries (too many LPM routes, too many ARP entries, Multicast...) then the chip will not be able to HW forward everything, and the CPU will be used.

You can issue that CLI command to have a view of your L3 resources usage. It's better to execute it during a normal day with a normal activity:

sh iproute reserved-entries statistics

Your limits is listed on the same table of the DS, and this output will also give you the limits. It could be useful to know what EXOS version you're using.

So, assuming you're within the HW capacity, your worst case throughput is 24*5 (slots used) = 120Gbps (or 240Gbps in Marketing math).

Hello,

Yes, we would like to see "how much traffic is currently routed".....

Reply