Cisco Support Community
Showing results for 
Search instead for 
Did you mean: 
New Member

6509E, four port 10G blade, buffers 16M per port

Looking at the specs for WS-X6704-10GE, it shows the port buffers at 16M per port.

This doesn't seem like very much for a 10G port.

Is this upgradable or related to how much RAM is in the blade?

VIP Super Bronze

6509E, four port 10G blade, buffers 16M per port

The blade buffer is not ugradable  If you are planning to use 10Gig blade, use the 8 port blade (X6708) instead.  The 6704 can not be used for VSL link, if you decide to use VSS in the feature.  The 6708 has a much higher buffere size (198 MB).

It also provide more qos features.


Hall of Fame Super Gold

6509E, four port 10G blade, buffers 16M per port

What supervisor card do you have?

10Gb line cards are supported only on the Sup720 and the Sup2T.

In addition to Reza's post, if you are planning to upgrade to Sup2T (soon or sometime in the future) you may want to reconsider as some blades (like the X6708) is not compatible with the Sup2T and the X6716 will need a hardware upgrade.

New Member

6509E, four port 10G blade, buffers 16M per port

We have the SUP720.

Why doesn't Cisco offer the higher performing baldes in smaller size?

We only need two ports and to get better performance and features, we have to spend more than double the money and will only be using two ports.

Cisco Employee

Re: 6509E, four port 10G blade, buffers 16M per port

x6704 has less buffer than x6708 as in theorical terms this is non blocking card. It is connected to the fabric via 2 x 20Gb fabric channels so 16M of buffer is deemed enough. x6708 on the contrary is a 2:1 blocking card (still 2 x 20Gb fabric channel) therefore bigger ingress buffers can help in case of congestion.

But this is not all...  due to the way buffering and forwarding is handled through the fabric the topic is much more complicated than this. We need to consider many other elements and not just rx/tx port buffers.

The key difference between the 6704 and 6708 cards is their Fabric ASICs with the latter having a completely different architecture which allows it to prevent those conditions which end up in fabric flow control events and in turn in drops which cannot be prevented even if much higher port ASIC buffers were available.

I cannot go too deep in terms of architecture (this place is not appropriate and I don't want to get you bored to death)... I just wanted to point out that port buffers are not the only element to consider.

In other word 6704 is a card which will work more than fine in the majority of implementations, but in case of heavily congested networks (with lots of mcast traffic which must be locally replicated by the replication ASIC) a 6708 would perform better even if it had the same buffer of the 6704 (provided that we only use for instance 2 ports on each card).



CreatePlease to create content