The gear I use in my test lab: A look at three Trident+ switches

Good enough can actually be so

Hardware-wise, these switches are almost identical. They are all based on Broadcom Trident+ Silicon and sport 9MiB of shared packet buffer. If you hammer these switches very hard and with particular traffic patterns you can elicit in any of them iSCSI microbursting issues. This is something overexcited storage admins worry about a lot.

I have caused iSCSI microbursting to occur on Trident+ switches in the lab if I place my servers one switch and my storage on another and use the 40GbE links to trunk between the two switches. Oversubscribe connectivity enough and you will absolutely run into it.

I have never seen this actually happen in production. The Trident+ silicon provides ultra low latency via cut through switching. Even though the packet buffers are only 9MiB for the whole switch the low latency access to the buffers means that microbursting is not an issue in practice unless your network is really, really oversubscribed.

Being a smaller buffered switch also helps guard against buffer bloat, which is its own problem, especially when there are TCP communications occurring with long round trip times. This isn't to say that the Trident+ switches are somehow ideal – I honestly don't think anyone has a solid handle on how best to size buffers for all workloads – but they've more than proven themselves for any workloads I've been able to throw at them.

I have only worked with one Trident+ switch – Supermicro SSE-X3348T – in a strongly multicast environment. It gave me no grief whatsoever. With more storage solutions using multicast, it is something that should be considered.

Choice is good

Looking at the above, we have three different approaches to from three different vendors for three different price points. This is fantastic; now we have options to really consider.

If I am in love with the idea of always having the latest features and protocols, then Cisco is going to be my only choice. Cisco's idea of "the life of the switch model" probably isn't going to be the full 10 or even 15 years that an SMB might try pressing a switch into service. On the other hand, when support does end on that switch it will much more modern than its counterparts. It's also almost double the cost of the Supermicro switch.

Dell's price point is the in-betweener, and the support plays out similarly. I expect that when the dell switch exits support, it too will have more features than the Supermicro. Like the Cisco this would help it stay relevant for longer, even past its official support lifetime.

Of course, simply being the cheapest option has its own advantages. With the Supermicro switch being almost half the cost of the Cisco I could simply buy another switch when the first one goes out of support, get whatever featureset was up to date at that point. I would end up paying about the same amount of money as the Cisco option, but end up with two switch lifetimes worth of security updates.

Of the options available, I personally prefer the latter. There are various reasons for this. The trite reason would be a personal bias towards the Supermicro SSE-X3348T. I have one, it's been remarkably good to me, so of course I find it easy to recommend to clients.

My choice isn't that simple. For the needs of my customers – SMB and top-of-rack networking – the Supermicro switch has all the features they could want. The price is cheap enough to shorten the replacement cycle and that keeps my security Mindspiders calm.

Which would you choose, and why? Answers in the comments. ®