We are looking to double the bandwidth on one of our circuits from 300Mbps to 600Mbps. We currently use a Cisco 7206VXR with an NPE-G1 card. These seem like very popular routers so I'm hoping a few people on this list have them deployed. If you or a customer have these deployed, how much bandwidth have you seen them handle? This will be handling dorm traffic at a college so it's mostly download. The 7206 handles our 300 Mbps circuit just fine, but we are moving it to our 600Mbps circuit. At peak we've seen the following numbers for that circuit:
30 second input rate 559982000 bits/sec, 55809 packets/sec
30 second output rate 55429000 bits/sec, 32598 packets/sec
267756984712 packets input, 333325152556755 bytes, 0 no buffer
This is the interface that connects to our provider. As you can see its almost all download traffic. Our ASR1002 handles it without a sweat but I'm a little skeptical of whether the 7206 will hold up.
This depends on multiple variables. The 7200 is a single-CPU platform where CPU can go sky-high when using features like ACL's, QoS, IPv6 and you name it.. Also, changing from IOS 12.4 to 15 increased
our CPU usage with another 10%+. Stick to the bare minimum of features you really need and you will be fine.
I have one but I never ran that much BW thru mine.
But the CPU usage is what will kill you.
Also the entire platform is rate for 1.8Gbs aggregated which mean
depending on which interface you have, and which bus they are connected
to, 900Mbps might be its limit.
We are looking to double the bandwidth on one of our circuits from
300Mbps to 600Mbps. We currently use a Cisco 7206VXR with an NPE-G1
card. These seem like very popular routers so I'm hoping a few people on
this list have them deployed. If you or a customer have these deployed,
how much bandwidth have you seen them handle? This will be handling dorm
traffic at a college so it's mostly download. The 7206 handles our 300
Mbps circuit just fine, but we are moving it to our 600Mbps circuit. At
peak we've seen the following numbers for that circuit:
30 second input rate 559982000 bits/sec, 55809 packets/sec
30 second output rate 55429000 bits/sec, 32598 packets/sec
267756984712 packets input, 333325152556755 bytes, 0 no buffer
This is the interface that connects to our provider. As you can see its
almost all download traffic. Our ASR1002 handles it without a sweat but
I'm a little skeptical of whether the 7206 will hold up.
I wouldn't expect a g1 to do much more than half a gig...
We're still on the 12.4 train. I do use an ACL with less than 100 entries which handle BCP38 and block a few bad actors and private IPs on the Internet. I will be moving the BCP38 ACL closer to the hosts before the upgrade so the ACL will be a bit shorter in the future. We won't be doing any QOS or IPv6 on it but it does take a full BGP table. I just need it to last another year or two out of it if possible. I believe this platform goes End of Support in Spring 2016.
Thanks for the link. When I looked at it, the PPS and bandwidth didn't really match what I see on my network so I'm curious to see what people are actually seeing. It looks like their test is done using very small packets (64K). Our traffic is mostly web with a lot of Video (netflix , Hulu, youtube, Flash etc) so we're dealing with a lot less packets that are much larger. Based on the numbers I posted, we' would be at the BW limit without even coming close the PPS limit (if we were running the traffic through the 7206).
The ACL is a recent addition and we can probably do away with it. I didn't notice a significant increase in CPU or drops since adding it. But we usually peak at about 200Mbps on this link. The full routing table is a must since we're dual homed.
We're still on the 12.4 train. I do use an ACL with less than 100
entries which handle BCP38 and block a few bad actors and private IPs on
the Internet. I will be moving the BCP38 ACL closer to the hosts before
the upgrade so the ACL will be a bit shorter in the future. We won't be
doing any QOS or IPv6 on it but it does take a full BGP table. I just
need it to last another year or two out of it if possible. I believe
this platform goes End of Support in Spring 2016.
yeah so you'll probably make it on a pure pps basis.
I do share the same thoughts as Remco. We've actually several NPE-G1 in production environments with full BGP feed. We saw a decrease in forwarding performance since 12.4T and up. We also recently disabled some features like netflow and ip inspection, which seemed relatively CPU intensive.
I do remember we were able to forward around ~700Mbps of 1500 bytes traffic with old IOS images and no ACLs.
Thanks for the link. When I looked at it, the PPS and bandwidth didn't
really match what I see on my network so I'm curious to see what people
are actually seeing. It looks like their test is done using very small
packets (64K). Our traffic is mostly web with a lot of Video (netflix ,
Hulu, youtube, Flash etc) so we're dealing with a lot less packets that
are much larger. Based on the numbers I posted, we' would be at the BW
limit without even coming close the PPS limit (if we were running the
traffic through the 7206).
so those pps numbers are worst case (small packet) but the acl count
/distribution and so on are going to impact what you actually get in the
downward direction.
in fact, the npe-g1 uses a BCM1250 which is a dual CPU unit but vanilla IOS
is not able to use the second CPU for packet forwarding. Unsubstantiated
rumour claimed that modular IOS (QNX kernel) could push about 1.6x the
throughput of vanilla IOS, as it was smp capable. Pity it was never released.
You don't necessarily need the full routing table for dual home, only
for outgoing load balance. You can have BGP, filter your routes away,
just leave a default gateway and still have dual homing. Your outgoing
traffic will work as if it were active-standby, though.
no, I meant modular IOS, not XR. This was an attempt to run a non
bare-metal IOS. The kernel was based on qnx (http://goo.gl/9RSwHn), and
cisco released it for the C6500 on the SXH and SXI code train. It turned
out not to be much of a success in the end - very little of use was
modularised, and it was canned after two minor code train releases. A bit
sad really, because it never had enough time to mature. It was never
released for any other platform. IOS-XE was a better implementation of non
bare-metal ios