Suggestion on networking reviews

Notice: Page may contain affiliate links for which we may earn a small commission through services like Amazon Affiliates or Skimlinks.

Railgun

Active Member
Jul 28, 2018
148
56
28
I appreciate I'm a new poster here so hopefully this comes across as constructive.

While I think this doesn't apply to many, would it be possible to do get performance metrics on switches? In particular latency.

It could be nice to see the characteristics of 100/50/25, etc to 10/1 interfaces and vice versa. Have they artichitected CT processing in these scenarios or SF. Do they have differences in latency vs frame size (probably with the latter). While getting line rate on a single interface with a constant stream of data isn't quite a stress test, how will it handle bursts? Do they have tunable buffers, etc. The Dell S5248F review for example gives an overall performance rating, but there isn't a single tested performance metric mentined in the article. What is that based on?

This may be beyond the intended purpose of the reviews, but perhaps go a bit more in depth into the actual capability, especially when you get into the higher performance catagory? Not sure how critical internal board layouts are to most folks.
 

Rand__

Well-Known Member
Mar 6, 2014
6,626
1,767
113
Not entirely sure what you are getting at, do you want built in performance measurements or external ones?
You could run smokeping to get latency values for example...

Or is this more a theoretical discussion to add performance measurement to reviews?
 

Railgun

Active Member
Jul 28, 2018
148
56
28
Theoretical to add to the reviews.

But to your point, external. The idea being to test through the switch, not within it. But does smoke ping have the granularity? In this context, we'd be talking ns and µs, not ms. Depending on the tools available to the reviewers of course. And depending on the method, if it's a software based only solution, chances are it won't be either 1) consistent or 2) that granular.
 

Rand__

Well-Known Member
Mar 6, 2014
6,626
1,767
113
Ok, so this is a "It would be great if" post, but no idea how to achieve that?
(Not trying to shoot it down, just trying to get context :))
 

Railgun

Active Member
Jul 28, 2018
148
56
28
I've got plenty of ways to achieve it. Whether it can or should be done by those that review it is the initial question. Whether anyone wants to see it is another. There are many ways to skin that cat. As I'm not doing the review, it's not up to me on how to achieve it.

But this is merely the suggestion, not the solution. :)

If it can be, I'm happy to throw in my $.02 or 2p or <enter currency here>.

For the record, I'm a network engineer with ~20 years in the financials (where this metric is just as important than raw throughput), so my want for this is probably vastly different than someone else's. And yeah, it's a nice to have, rather than need to have.
 
  • Like
Reactions: Patrick

Rand__

Well-Known Member
Mar 6, 2014
6,626
1,767
113
Well I think if you can provide an easy, repeatable, consistent solution that adds value to reviews then they will be all for it;)

The more complicated and time intensive it gets, the less likely you are to see something like that.
 
Last edited:

Patrick

Administrator
Staff member
Dec 21, 2010
12,511
5,792
113
Hi @Railgun

We are working on this a bit, but for lower-end reviews.

The big challenge is that the cost of doing this for anything >10GbE gets to be a lot. For 100GbE even a two-port test fixture with snaked cables costs a ton.

We tried doing T-Rex, that ran into issues by the time we got to 25GbE. Spirent and ixia for 25GbE and higher are very expensive. If we had something easy for T-Rex, then we would use it, but we spent thousands of dollars building a test box dedicated to T-Rex and ran into issues at 25GbE while also not having great 10GE results.

The other challenge is just the setup time to get something reliable enough to be useful.

I think that @Rand__ has the basic idea here. At some point, there is always a LOT of stuff I want to do. On the other hand, doing all of those things on every review costs a ton. Rohit and I are basically focusing the current series on just getting stuff out there. Right now, the number of sites that do anything around switches is almost zero. While it would be more valuable to our readers if we could do everything, right now we have basically $0 ad revenue from networking vendors (I may be wrong since all ad sales are external.) So we are at the point where we are trying to do the best we can subsidizing the reviews.

I know that is not a great answer. At the same time, it is what we have right now.
 

Railgun

Active Member
Jul 28, 2018
148
56
28
I appreciate that. I’m happy to pop some suggestions that make it a bit more viable. Hell, I’d even be happy to help support such a setup.

Agreed that when you start looking at Ixia, Spirent and the like…somewhat hardware based solutions, it gets expensive. I did exactly this early last year for a particular use case at just 40Gb. But in any case, it’s really just the software they use. The optics for 100Gb aren’t bad ($100 for SR4 QSFP28s). NICs on the other hand…. However, if you‘re not going to test latency at line rate, you don’t necessarily need to have 100Gb in the whole chain. I’ll explain our method that worked in a later post here when I have more time.
 
  • Like
Reactions: Patrick

Patrick

Administrator
Staff member
Dec 21, 2010
12,511
5,792
113
I think Spirent also uses FPGA-based cards. We actually have a C100 in the lab now for SFP+ but I am unsure if we will use one long-term.

On the 25GbE+ the few hundred for optics is not the issue. Getting cards that work with T-Rex / DPDK and such that we can use for testing, plus getting testing down to something easy to execute is the main issue. If this was a $5000 problem to get good 25GbE/100GbE testing we would have already done it. In fact, I think we spent more than that already.

The other, not insignificant, the challenge is that there are two ways to test:
1. We set up some basic test methodology, just configure that layout and then test
2. We go in-depth to enable features/ check firmware versions

The first gives us a least common denominator. The second point is what we would need to do in order to get top-end performance. In some devices, especially think routers/ firewalls, the second is even more important. Performance varies widely based on those combinations. While I prefer #2, it also means that not only do we need a system to test with, but we also need to know the best setups for each vendor and device.