@Hc-angie-chen , when you look at the diagram @MHM Cisco World provided, the way it's diagramed, there doesn't appear to be much difference between using StackWise vs. SVL; in this diagram, there's not.
The diagram, especially, shows core and distro pairs (?) for both StackWise and SVL. In a single pair relationship, StackWise appears to offer about the same capabilities and capacity as SVL, but you need to consider the platform capabilities/capacity being used and whether we're doing L2 or L3.
With L3, such as between core and distro, SVL, does as mentioned in the diagram "Main goal is to simplify Distribution or Core layer". Consider, L3 between devices can use multiple paths, all paths, and shortest path. I.e., what does SVL actually offer compared to just using L3? Actually, not much. It halves "devices", but makes their configuration more complex. Possibly it reduces reliability, as possibly SVL software, being more complex, may be more likely to have a bug. It reduces some options, such as using different peer devices. Frankly, for a pair of core devices, like in the diagram, we often decided to keep them independent L3 devices. (BTW, when I described SVL operation, for its data path usage, it much mimics L3, as L3 peers don't generally send traffic sideways, unless it's the only way to reach the destination.)
StackWise or SVL, are very nice for L2. It avoids all the issues that come with spanning tree (regarding using all links using Etherchannel between multiple physical switches because they are now just one logical switch). However, since there's one logical device, logically, traffic might move sideways between peers.
Consider using 6513, with sup2Ts. (I'm using old technology, as I'm very familiar with them.) You have 11 line card slots, each providing 80 Gbps, duplex, which is non-blocking with the 2Tbps fabric. If dual peers are logically one device, how to you interconnect them to also provide for 4 Tbps? With SVL, the peers do not want to pass data between themselves, so if you design for that, it's a non-issue. With StackWise, you try to provide the necessary bandwidth because the peers do not try to avoid using each other.
Again, with just two peers, for StackWise, other than an extra hop, if you unnecessarily use a peer, there may be enough ring bandwidth, that bandwidth is not oversubscribed (as it may be with a larger number of stack members).
However, traditionally, StackWise switches do not have the hardware resources of the SVL level of switches, i.e. they may have performance issues unrelated to the stack, itself.
Again, going back years ago, Cisco had the 3750G with 48 copper gig ports (plus uplinks) and the 4948 with 48 copper gig ports. Hmm, both has 48 copper ports, so why what the 4948 so much more expensive?
Well the 3750G fabric was on 32 Gbps (enough for 16 gig ports) while the 4948 fabric was 96 Gbps. So as many found out (peruse old forum posts), 3750Gs didn't do well in distro or core roles, or even server edge roles, if they were "busy".
Cisco then came out with the 3750E, which also had a non-blocking fabric, and PPS to support wire-rate for its 48g copper ports; much like the 4948. So, it was not just as good as the 4948? Better, as it supported stacking, right? Well, no, because it didn't have the buffer capacity of the 4948. so burst oversubscription of edge ports would often lead to drops, not seen on a 4948. Same problem on the 3750-X and the 3650 and 3850 too. Again, if desired, peruse forum posts. BTW, the 4948 wasn't stackable, but eventually, other members of the Catalyst 4K series, did obtain VSS.
To recap, both StackWise and SVL offer about the same redundancy and L2 features, but SVL somewhat mimics how L3 routing would operate for path selection. Basically, it considers using its peer as an extra high cost hop, so it generally avoids it. If you have lots of east-west peer traffic, VSS/SVL is worse than StackWise, yet even StackWise can run into peer<>peer(<>peer. . .) bandwidth issues.
If you have very high bandwidth requirements, between devices at the same level, your best performance option is using a single device that supports the bandwidth requirement and port density (reasons for existence of large physical devices - consider something like Cisco's CRS devices), or using a logical multi device fabric, Cisco's APIC/ACI, otherwise, you use a hierarchal topology, but care needs to be done supporting redundancy.
BTW, in another vendor's product portfolio, which I've used, they have stackable switches than can use high bandwidth local stacking, or "ordinary" links for non-local stacking, or both, concurrently. I believe, they treat both, internally, much like a multi-switch fabric, supporting one logical switch. Logically superior, I believe, to either StackWise or VSS/VSL, but a later technology development than Cisco's. Further, especially with VSS/VSL, it was possibly designed for devices never designed for this. So, not surprising it might be logically superior, however, also possibly, its devices didn't seem as operationally solid as Cisco's. (Many of us mention issues with Cisco devices, but compared to many Brand X devices, Cisco devices often work as documented, and when they don't, Cisco actually provides updates so that they do.)
