Non-blocking Cisco switch recommendation for iSCSI data center

I've read a myriad of docs/articles, and heard many varied opinions on what "non-blocking" operation really is, but still haven't settled on a solid answer.  Specifically, I'd like to compare two current Cisco switches, and get feedback on how they compare - the Cisco 2960-S -, and the 2360 -

Cisco positions the 2960-S as an edge/access switch (and clearly it has such features as PoE, etc. that the 2360 doesn't), while the 2360 is positioned as a data center switch.  The data sheets indicate that the 2360 is non-blocking (the 2960-S does NOT indicate this), but yet their forwarding rates are identical at 88-Gbps.  I have two real questions here:

1) what indicates if a switch is truly non-blocking?  Use these two switches in comparison to explain.

2) I can't really see much difference between these two products from a throughput, traffic forwarding perspective, and yet the 2960-S 48-port w/ 10-GE uplinks lists for $6500, while the 2360 48-port lists for $8700.  I see that the 2360 supports something called "dynamic buffer allocation", but the datasheet hardly spends any time on this.   From a data center performance perspective, why is the 2360 a much better solution than the 2960-S?

Thank you, and reference links/docs are always appreciated.

Who is Participating?
Don JohnstonConnect With a Mentor InstructorCommented:
1) It's not that there's confusion, it's that there's no definition. And while the backplane is 8gig short of meeting that definition, it would only be an issue if all 48 ports were fully utilized (in full duplex). BTW, another definition of non-blocking is the backplane is greater than the sum of all the ports (without consideration of full duplex). Which in this case, you have more than enough bandwidth in the backplane.

2) You're trying to apply logic to a marketing issue.  That's just not going to work. :-)
For a top of rack switch, being able to move the data to the end of rack is VERY important. Hence the 40gig of uplink ports. The additional buffer capability is just icing on the cake.

As for which one is best? That's going to depend on where it's positioned. If this is going to be a basic access layer switch, the 2960 should do fine. If it's going to be top of rack switch in a small rack, then the 2360 is a better choice.

Ultimately, it depends on traffic flow. How much traffic is this switch going to see and how much of that traffic is destined out the uplinks.
Don JohnstonInstructorCommented:
1) These really is no "official" definition of non-blocking. Some say if the backplane speed is greater than the combined speed of all the ports, that makes it "non-blocking". Another definition would be if the backplane speed is greater than all the ports AND there is a path for that traffic (uplinks).

The 2360 that has 48 1gig ports, you would need a backplane of 48gig x 2 (full duplex, right? ;-)) which means your would need a 96gig backplane. Okay, that's close enough. You've also got four 10 gig ports so you've got almost enough bandwidth to carry the traffic from all the gig ports. So I would say that it's close enough to be called "non-blocking".

The  2960 has the same backplane but lacks the uplink capability.

2) A) The 2360 can have four 10 gig ports while the 2960 can only have two 10 gig ports.  and B) two words: "Data Center" Each word is worth about a grand apiece. :-)

cfan73Author Commented:
Hey Don - thanks for the response...  some follow-up:

1) See, if the backplane were the issue, then neither of these switches would qualify (based on the full-duplex math) - and Cisco claims the 2360 is non-blocking and the 2960-S isn't, even though their backplane forwarding is identical.    I read something elsewhere where the backplane BW isn't really the issue at all, but rather a "number of inputs vs. number of outputs" on the switching fabric, etc.  Clearly, there's a lot of confusion on this topic...

2) Funny, but there has to be more to the differentiation between these two families, seeing that Cisco is drawing such a clear positioning for each.  I'm stuck on the "dynamic buffer allocation", which is pretty much the only difference between the two (beyond the additional 10-GE uplinks), so I'm trying to find out more about this, and how it would be a better fit for iSCSI operation in a data center fabric.

Thanks again
kevinhsiehConnect With a Mentor Commented:
My guess is that your environment is relatively small, or we would be talking about larger chassis or at least stackable switches. Let's put things into some perspective. Since we're talking iSCSI, your network is going to be oversubscribed to begin with. I have seen iSCSI storage systems with up to 8 1 GB network connections. If on a 48 port switch you have 40 servers with 1 GB of bandwidth attached to a san with 8 GB of bandwidth, does it matter if the switch can move 32 GB or 48 GB before the switch starts to drop packets? No, because you are limited to 8 GB max on the storage device, and you can't more iSCSI traffic on the switch than that. As long as your switch meets a certain minimum performance level, and the 2960-S does, you are going to be fine.

I use two 2960-S switches in my environment where my servers and EqualLogic SAN are directly connected to the two switches. The switches are trunked together, and I have 2 GB trunked uplink to my 6500 core switches. I have about 85 volumes online, with 480 iSCSI connections to the SAN, and it is rare that any single port on any of my switches reaches even 10% utilization. I don't even use jumbo frames on my network. iSCSI rides on TCP, so it can handle dropped frames and packets if they did occur. No need to overthink this purchase. Get the less expensive switch.

Question has a verified solution.

Are you are experiencing a similar issue? Get a personalized answer when you ask a related question.

Have a better answer? Share it in a comment.

All Courses

From novice to tech pro — start learning today.