Broadcom Takes The 4x4 Road For 802.11ax

[sm00thpapa]

Definitely won't be buying any Broadcom product that's for sure.

 

[sfx2000]

My guess is despite the fact the standard won't be finalized until 2019, manufacturers will still hit the market with products somewhere in 2018, people will pay 699$ for it despite @thiggins numerous "stop paying to be a betatester" warnings, and SNB forums will be filled with unhappy users.

Hehe - remember the 802.11n (draft, pre-draft, lets ship something, and finally a release - and then a bug fix release, e.g. 2.0)...

 

[sfx2000]

They keep trying to increase the speed over wifi but still leaving the wired ports at 1 gig. When are these companies gonna start making routers with 10 gig ports or higher.

10GBe is still pretty expensive per port... on switches and clients.

N-Base-T - much more real in terms of cost - not much more than 1GBe.

 

[pete y testing]

They keep trying to increase the speed over wifi but still leaving the wired ports at 1 gig. When are these companies gonna start making routers with 10 gig ports or higher.

one already has the netgear x10 , only 1 port and prob designed for a NAS

 

[sm00thpapa]

10GBe is still pretty expensive per port... on switches and clients.

N-Base-T - much more real in terms of cost - not much more than 1GBe.

Really? How much could a 1/2 inch by 1/2 inch square cost coming out of China?

 

[sm00thpapa]

Hehe - remember the 802.11n (draft, pre-draft, lets ship something, and finally a release - and then a bug fix release, e.g. 2.0)...

Yep 2.0 N Draft was out for like 2 years before final. What a joke.

 

[sfx2000]

Really? How much could a 1/2 inch by 1/2 inch square cost coming out of China?

It's not the physical port as you suggest

N-Base-T leverages into the 1Gbe platforms already out there - so it's much easier and cheaper to deploy than 10Gbe...

I do see 2.5Gbe/5Gbe as being much more interesting and closer to the consumer market... and more than good enough with 11ac (Wave2) and 11ax/11ad deployments.

Recall, I'm an engineer, not a marketeer...

 

[sfx2000]

one already has the netgear x10 , only 1 port and prob designed for a NAS

That port on the X10 raises the price quite a bit, and then one still has to factor in the cost of the SFP+ adapter... and most home users won't use it. Seriously...

 

[pete y testing]

and most home users won't use it. Seriously...

true , but if its there some will still buy it

i do agree with other comments that with the likes of this new standard ethernet may start to become a thing of the past for most none commercial use

 

[sfx2000]

MU-MIMO uplink is part of 802.11ax, which is still in draft form. But, like 11ac, all features are not implemented in first parts.

I think UpLink MU will likely fall by the wayside... not much benefit, and it seriously complicates the UL link for OFDMA (I worked on 802.16e/16m) - LTE decided to just not do that - keeping the uplink simple with pre-coded OFDMA, which they call SC-FDMA...

There is a thing called - just good enough...

There's still benefit with 802.11ax in that the MAC layer is now scheduled - and this is a good thing, seriously, it is, for capacity and quality of service. In a non-mixed mode 11ax BSS, performance should be very good... legacy devices, they'll suffer, and that includes all 802.11 a/b/g/n/ac once more 11ax is in place - and that includes adjacent BSS's... 11ax should be neighbor friendly, but as more neighbors adopt 11ax AP's and clients.. the future will be clear.

One of the interesting trends in LTE-Advanced is the concept of Massive MIMO - where the Tx Chain at the eNodeB might have 64/128 spatial streams - and the client can combine as needed based on signaling - the uplink is still Single Carrier/Precoded 2-stream OFDMA, and it works - actually very well...

In the SOHO market for 802.11 - I could see AP's at 8*8:8 easily enough on a 80MHz channel... push even more bits, shannon is a limit, but we get more and more clever all the time on shoving those bits into the channel.

Why 80MHz? - because of regulatory constrains on a global basis - 160Mhz channels are hard as many regions do not have contiguous channel space to do it, and 80+80 isn't cutting it - as we see with 802.11ac Wave 2.

 

[RMerlin]

Hehe - remember the 802.11n (draft, pre-draft, lets ship something, and finally a release - and then a bug fix release, e.g. 2.0)...

At least draft N devices weren't too far away from the final release. Remember the 56KFlex / v90 days?

 

[RMerlin]

Really? How much could a 1/2 inch by 1/2 inch square cost coming out of China?

You need a processor able to handle the packets coming in and going on at 10 Gbps. If you have four ports, then the processor needs to be powerful enough to handle 10 Gbps * 4 ports * 2 (full duplex), for a total of 80 Gbps of traffic switching. A 5$ SoC won't be able to process that amount of traffic... Add to that you need memory buffers to be 10x larger than on Gigabit Ethernet to keep up with that.

A network switch isn't just an electrical connection, ASICs are involved to actually process the data.

So don't expect home routers to offer more than one single 10 Gbps in the near future, maybe two at most. 2.5 and 5 Gbps sound more likely.

 

[sfx2000]

Remember the 56KFlex / v90 days?

Indeed - I was on the US Robotics side there... I still have the v.everything Courier sitting on the shelf in my home office.

 

[sfx2000]

You need a processor able to handle the packets coming in and going on at 10 Gbps.

Most ARM's, and many Intel/AMD x86-64 chips - they're not going to handle 10Gbe for routing in SW.

Even with 1Gbe - software routing is generally clock limited - it's packets per second, and this is clock speed on the SoC... linux/bsd/mac/windows - all are generally good at routing these days for SW...

 

[pege63]

I do see 2.5Gbe/5Gbe as being much more interesting and closer to the consumer market... and more than good enough with 11ac (Wave2) and 11ax/11ad deployments.

For me 2.5Gbe/5Gbe would be moore interesting sence i have most off my clients on the cable side, but anyway most off them are 1Gbe.
But still it would require switches who could handel the 2.5Gbe/5Gbe speed.

 

[pege63]

Remember the 56KFlex / v90 days?

Those were the days when you had USRobotics modem i still have them i my basement from the first 9600 bit/s up to the V92 standard, every one of them.

 

[thiggins]

SFX: Why does UL MU not provide much benefit?

 

[sfx2000]

SFX: Why does UL MU not provide much benefit?

Similar to Wimax - the sounding and ranging is a lot of work computationally - and since the traffic mix is largely asymmetric, e.g. much more downlink traffic in typical use than uplink...

And just like MU downlink, the traffic has to be a good fit for the scheduler, otherwise it's actually better to use SU instead.

MU is hard - QCA is best at it, and I totally understand the challenges that Broadcom (and others) have had with MU-MIMO on 11ac.

 

[thiggins]

since the traffic mix is largely asymmetric, e.g. much more downlink traffic in typical use than uplink...

A few years ago this was true. But with more people doing live streaming, uploading selfies from phones with huge megapixel capacity, upload bandwidth use is increasing.

At any rate, thanks for the insight.

 

[sfx2000]

A few years ago this was true. But with more people doing live streaming, uploading selfies from phones with huge megapixel capacity, upload bandwidth use is increasing.

At any rate, thanks for the insight.

Based on current stats - the downlink is still predominate - uplink from certain users perhaps, but it's a blip on the big data picture...

We want the uplink to be fast and relatively uncomplicated. Consider the population of devices...

This is why MU-uplink in 11ax is not needed... at least not in the next few years, and even then, things likely will change.