Catalyst 9300 StackWise: 320, 480 and 1T Bandwidth Explained

Cisco's Catalyst 9300 naming is confusing precisely because the "StackWise" numbers describe different hardware generations, not different speeds of the same design. The official Catalyst 9300 StackWise System Architecture white paper clears it up: the number after StackWise (320, 480, 1T) is the full-ring stacking bandwidth your model can deliver when Switch Rollover Protection (SRP) is enabled. This article decodes the white paper's table so you can size a stack correctly and stop guessing what your uplinks can actually carry.

The confusion is understandable, because three different things in the Catalyst 9300 family are all described with numbers that look like bandwidth. There is the per-ring speed, the number of rings, and the resulting full-ring total - and only the last of those appears in the product name. Once you separate the three, the model selection question becomes arithmetic rather than guesswork.

The StackWise Generations in One Table

• Catalyst 9300X (StackWise-1T) — 6 rings at 90 Gbps each = 540 Gbps full ring, doubling to 1 Tbps with SRP.
• Catalyst 9300 modular uplink models (StackWise-480) — 6 rings at 40 Gbps = 240 Gbps, doubling to 480 Gbps with SRP.
• Catalyst 9300 fixed uplink models (StackWise-320) — 4 rings at 40 Gbps = 160 Gbps, doubling to 320 Gbps with SRP.

The default back-plane stack interface speed for Catalyst 9300 and 9300X is 480 Gbps — note that default applies across the family even though the full-ring maximums differ.

Reading the table properly takes a moment, because the arithmetic is not uniform across the rows. The fixed-uplink 9300 has four stacking rings running at 40 Gbps, so an unbroken ring carries 160 Gbps. The modular-uplink 9300 keeps the same 40 Gbps per ring but has six of them, giving 240 Gbps. The 9300X changes the per-ring speed to 90 Gbps and keeps six rings, giving 540 Gbps. In every case the product name reflects the number you get after SRP doubles it: 160 becomes 320, 240 becomes 480, and 540 becomes 1 Tbps in the marketing round-up.

Why "the number doubles with SRP"

The step that trips people up is that the ring is not the same thing as the usable bandwidth, and the difference is entirely about failure behaviour.

A physical stack is wired as a ring: the last switch connects back to the first. Traffic between any two members can travel either way around the ring, but the ring as designed is provisioned so that a single cable break does not partition the stack. Without Switch Rollover Protection, a break in the ring leaves you with a chain, and the traffic path between some members is lost. With SRP enabled, the stack detects the break and rolls the ring over onto the surviving path, turning the broken topology into something that behaves like a full-duplex path. That rollover is why the effective bandwidth doubles: 160 to 320 Gbps on fixed-uplink 9300s, 240 to 480 Gbps on modular-uplink models, and 540 Gbps to 1 Tbps on the 9300X. The detail worth remembering is that the doubled figure is a capability rather than a measurement - you only realise the benefit when a cable actually fails, which is why it is easy to overstate how much headroom a stack really has on a normal day.

The practical consequence is that "SRP enabled" is not a cosmetic setting you leave at defaults and forget. It belongs on your build checklist next to STP root and NTP, and it belongs in your verification output, because a stack that silently lost SRP after a firmware upgrade behaves normally right up until the moment a cable is disturbed.

Ring topology, cabling and what breaks

Catalyst 9300 stacks are built with dedicated StackWise cables, not with uplink ports. That distinction matters for two reasons. First, stacking traffic never competes with uplink traffic for the same physical media. Second, the stack survives an uplink failure with the full ring intact - a failure domain that is invisible to the uplink design.

The cabling rules worth internalising:

• Every member needs both stacking ports populated to form a ring. A stack built in a chain (one cable per switch) still works but loses the redundancy the design intends.
• Keep the cables within their supported length for the stacking speed in use. Long runs between closets are not a StackWise use case; that is what StackWise Virtual over uplinks exists for.
• Power stacking is a separate mechanism from data stacking. It lets one member supply power to another and has its own configuration and its own failure modes, so do not assume a healthy data stack implies a healthy power stack.
• Document which physical port of which switch connects to which. When a stack misbehaves at 2 a.m., the first question is always "which cable is that?"

Reverse the thinking for a moment and consider the failure modes: a single cable cut leaves the stack running with SRP providing the rollover path; two cable cuts at different points can partition the stack into islands, which is exactly why cable diversity across the members matters; and a member failure removes capacity but not connectivity. The bandwidth number in the product name is the budget you have to cover the third case.

What SRP Actually Does

Switch Rollover Protection makes the stack resilient to a broken or removed StackWise cable by rolling the ring over onto itself, turning a half-duplex ring into what behaves like a full-duplex path. That rollover is why the effective bandwidth doubles: 160 to 320 Gbps on fixed-uplink 9300s, 240 to 480 Gbps on modular-uplink models, and 540 Gbps to 1 Tbps on the 9300X. If an SRP-capable stack is running without SRP enabled, you are deliberately operating at half the bandwidth the hardware supports.

How much bandwidth does your stack actually need?

This is the question the product naming is trying to help you answer, and it is best approached as three numbers rather than one.

1. Aggregate uplink capacity - add up the uplink bandwidth of every member. Four fixed-uplink 9300s with two 10G uplinks each is 4 × 2 × 10 = 80 Gbps.
2. Worst-case redistribution - decide how much of that aggregate has to survive the loss of a single member. In a stack where every member's uplinks are in the same LACP bundle to the same upstream pair, losing one member does not concentrate traffic on the others, which relaxes the requirement.
3. Stacking budget - compare against the full-ring number for your model. 80 Gbps sits comfortably inside StackWise-320's 320 Gbps with SRP enabled, or 160 Gbps without.

Run the arithmetic on a denser design and the answer changes quickly. Eight 9300X members with four 25G uplinks each is 8 × 4 × 25 = 800 Gbps of aggregate uplink. That exceeds the 540 Gbps full ring of a StackWise-1T stack without SRP, and fits only inside the 1 Tbps figure with SRP enabled. A stack of that shape on fixed-uplink 9300s would be a design error regardless of how the uplinks are bundled, because the backplane becomes the bottleneck long before the uplinks do.

The sizing rule of thumb follows directly: the stacking bandwidth should exceed the aggregate uplink capacity of the members, otherwise the stack backplane becomes the bottleneck during failure.

StackWise vs StackWise Virtual

Physical StackWise uses dedicated StackWise cables to build a single control plane across up to several switches in the same rack — the model behind the 320/480/1T numbers above. StackWise Virtual is a different technology: two switches are virtualized into one logical switch over their uplink ports, which allows the pair to be split across longer distances such as different rooms or closets. The white paper and Cisco's StackWise Virtual documentation clarify that physical stacking targets access-layer density while StackWise Virtual targets distribution and core roles where two-node pairs over uplinks make sense. Our Catalyst 9300 StackWise vs VSS vs StackWise Virtual comparison walks through the deployment choice.

A useful way to hold the distinction: physical stacking scales outward - more members, more access ports, one management IP - while StackWise Virtual scales upward and outward across distance, at the cost of consuming uplink ports for the virtual link and being strictly a two-member design. If the requirement is "eight switches in one closet with one control plane", that is StackWise. If it is "two switches in different rooms acting as one", that is StackWise Virtual.

Choosing the Right Model for the Stack

Match the model to the workload: fixed-uplink 9300s (StackWise-320) are fine for general access; modular-uplink models (StackWise-480) suit access blocks with heavy video or wireless backhaul; the 9300X (StackWise-1T) is for the densest access and distribution roles where the stack also carries inter-block traffic. Sizing rule of thumb: the stacking bandwidth should exceed the aggregate uplink capacity of the members, otherwise the stack backplane becomes the bottleneck during failure. For the full picture of the older Catalyst stacking options and VSS, see our StackWise vs VSS vs StackWise Virtual comparison.

Verifying the stack before you trust the number

Whichever generation you deploy, the configuration is only half the story. The other half is what the stack reports about itself, and there are a handful of commands worth putting in your post-install checklist:

show switch
show switch detail
show switch stack-ports
show switch neighbors
show stack-power
show run | section stack

Look for three things in that output. Which member is the active and which is standby, and whether the member numbering matches your documentation rather than the order the switches happened to be powered on. Whether every stack port reports "OK" on both ends - a port that is "Down" or "Not Present" tells you the ring is incomplete even though the stack may still be forwarding. And whether the reported stack interface bandwidth matches the model you expected; a stack of 9300X members that reports the family default rather than the full ring is a configuration or cabling problem, not a marketing one.

Common misconceptions to drop

• "480 Gbps is the family bandwidth." It is the default back-plane stack interface speed, not the full-ring maximum for every model. The 9300X reaches 1 Tbps with SRP.
• "The number is per-direction." The published figure is the full-ring total; treat it as the pool the whole stack shares rather than a per-member allowance.
• "SRP is optional polish." Without it you run at half the advertised bandwidth and lose the rollover protection the ring is designed for.
• "Any two 9300s can stack." Models differ in StackWise generation, and the members in a stack need to be compatible; confirm the model pairing against Cisco's stacking guidance before you order hardware.
• "A stack is a replacement for a proper uplink design." The ring carries member-to-member traffic. If the uplinks are oversubscribed, a faster ring will not fix it.

Planning checklist

• Count the uplinks first, then pick the StackWise generation that covers their aggregate - not the other way round.
• Decide whether SRP is enabled, and record the decision, because it changes the number you are sizing against.
• Confirm the stack members are a supported combination for the generation you chose.
• Design uplink bundles so that losing a single member does not force its traffic through an already saturated path.
• Verify after install with show switch stack-ports and keep the output in the build documentation.

None of this requires trusting the product name alone. The three rows of the table, the SRP doubling rule, and the aggregate-uplink comparison are enough to size almost any Catalyst 9300 stack correctly, and to explain the answer when someone asks why the 9300X was specified for one closet and not the next.

原文链接：https://www.cisco.com/c/en/us/products/collateral/switches/catalyst-9300-series-switches/white-paper-c11-741468.html