Marvell Has Low-Cost 10GbE For Home Users But Demand Is Still Low

Status
Not open for further replies.

InvalidError

Titan
Moderator
I don't see how moving PHY to software makes much sense as the hardware implementation is far more power-efficient and space-efficient than a software-based equivalent with the associated much stronger CPU requirement. The biggest overheads for power and area are the ADCs, DACs and analog amplifiers which are still required either way.
 

thundervore

Distinguished
Dec 13, 2011
1,030
1
19,460
We need 10GbE in homes now!!!! I have a 1Gb network at home and it crawls when im accessing my server trying to back up a computer while pushing content to 3 TVs and torrenting. My switch get abused that's connected to the server.
 

dstarr3

Distinguished
Well, there just aren't that many homes yet that have ethernet in the walls, let alone CAT6 or CAT7. My house was built ten years ago and I'm lucky enough to have ethernet jacks in every room, but it's CAT5e, so I'm not going to break the 1Gb barrier anytime soon. So if you want a house with CAT6 or CAT7, you either need to buy a very recently built house, or have it installed in your own house.

That, and American internet is pathetic and simply no one has the need to wire their house up with anything more. It's just reserved for enterprise and the tech enthusiasts at this point. The enterprise sector has no interest in budget components as hardware cost isn't a concern, and the tech enthusiasts are a very small niche that never will represent a significant market space.
 

Vatharian

Distinguished
May 22, 2009
90
0
18,630
I've read the article couple of times, to make sure I didn't missed anything, or if there was some line I missed. That's load of bullcrap! They're looking at the effects and completely mistaking them for the cause.

First, Thundrbolt 3 has absolutely NOTHING to do with networking. Number of devices supporting it is minuscule, and while they have 'wow' factor for using really extreme interface, they are far from common.

Second, high speed WiFi is the main factor pushing for 10GbE at home environment. What's the point of 1.7 Gb/s+ connection, when all you can do is mere 1 Gb/s, because cables? Add a NAS, and here you go, big damn bottleneck, you'd all be the same while using 802.11n. MIMO? What for?

Third, normal home users would love to have 10 GbE, if there was absolute lack of switches. Give us affordable 8 or 12 port 10 GbE switches. Consider routers presenting 6x 10GbE LAN ports (vendors that put 4 RJ45s on routers are the root of all evil!). You want to see good penetration and have cheap solution? USE IT! Soon, EOLed and deprecated enterprise equipment supporting 10G will start popping on auction sites, and it's selling like hot cakes - It's the same when 1G saw big adoption spike. But times have changed, and it's not ideal solution for many. Again, want sales - give us something we can buy!

Next thing is, quit yapping, that there is lack of content that would benefit from 10 GbE at home. Now, mostly true. See 10% market share, and grab popcorn and watch what happens.

There is considerable number of people, who still sport 100 Mb/s networks at home, because when they were lying down the infrastructure, 1G was prohibitively expensive. Their lousy low quality Cat 5 cables won't support 10GbE, and not even 2.5 or 5G. But... it's around the time, when they are renovating their homes - there is great sales potential there.

Next thing, going completely wireless is really the last thing we want. Forget about security (or lack of it) for a moment, but in most places where more than 2-3 famililes live close enough, 2.4 GHz is completely cramped, and 5GHz is starting to see congestion problems. Then, secure, stable and fast wireless is NOT plug and play.

Finally - we don't want half-baked and half-assed solution like 2.5 and 5G. It's a waste of effort, and looking for a niche that doesn't exist. This technology is already outdated!
 

InvalidError

Titan
Moderator

People who are still on 100Mbps on their LAN would be more than happy with 1Gbps for the foreseeable future and most aren't going to be looking at 10G any time soon. Also, even cat5 can handle 10GBase-T over shorter distances with the newer variants stretching copper even further due to the number of companies preferring to pay for more complex PHY chips over having their offices rewired. That's how 1GBase-T which worked on cat5 ultimately displaced the cheaper 1GBase-TX standard which required cat6a.
 

CRamseyer

Distinguished
Jan 25, 2015
425
10
18,795
Let me change you mind on some of those points.

Thunderbolt 3 devices and 10GbE with a NAS do share some target markets that benefit from high-speed connected storage. Think video editors and photographers. Thunderbolt 3 will gain market share next year when Intel brings the technology on the CPU die as announced just a few days ago. Everyone interested in 10GbE for the home would rather have a high-speed Ethernet connection but TB3 will be easier since we don't have the switches ready.

2.5GbE is powerful enough for the new WiFi Devices. Both it, and 5GbE, will operate over Cat 5e.

EBay is full of low-cost surplus 10GbE switches that were ripped from data centers but most sound like a jet plane. Most people can't live with the sound of a spooled up turbo charger sitting on the desk. Asus had the right idea with the XG-U2008 but it only gives users access to two 10GbE ports.
 

InvalidError

Titan
Moderator

How many people are going to need more than a pair of 10G ports on a home router/switch? One port for the NAS, one port for the workstation/main PC sees the bulk of all significant use, most other devices in a typical home aren't going to need more than 1Gbps on a frequent enough basis to justify the additional cost and power any time soon. Personally, I prefer keeping my backups offline using an external USB3 HDD where malware like WannaCry can't touch 'em.
 

dstarr3

Distinguished


The vast majority of video editors and photographers don't have the cash to have all of their data stored on SSD arrays. Rather, they're still using hard drives. Huge capacity, low cost per gig. And if all your data is stored on HDDs, you're not going to see a performance boost by switching to 10GbE. A hard drive at full sequential speed just barely fails to saturate even 1GbE.
 
Wow talk about... whatever.

Here is the scoop, 10GbE is annoyingly expensive for good reasons. Currently you need special cables to go any reasonable distance due to the way the signal is handled. Using RJ45 copper of any kind gives pretty bad distance without an expensive and unreliable SFP unit. It's unreliable because the power required to drive a 10GbE signal down a piece of copper is atrocious at current lithography standards. There are newer 14nm chips (both PHY and SFP) on the horizon that promise to reduce this load significantly and thus reduce the costs. Also the author confused something, PHY's and SFP's are essentially the same thing. PHY's drive the signaling for a set of ports, SFP's are a modular-PHY that drives the signal for a specific cable.

As for 10GbE in the home, not happening anytime soon. There is almost zero demand for it, only a handful of professionals wanting to tinker around with it and the physics behind it make it cost prohibitive to do with copper. The 2~5 GbE is far more likely to become commonplace due to it's much lower electrical requirements.

This is the same reason processor clock speeds have dramatically slowed their growth, the physics start to get in the way.
 

terr281

Distinguished
Dec 22, 2008
261
0
18,790
In my region of the Southeast USA:

1. People understanding the basic concept of wireless security is a difficult find.
2. Most don't know that connecting their "cool" laptop to a wire, instead of working via wireless, is faster.
3. Most homes, unless in high value subdivisions, may have RG6 in the rooms as the only (useless) option.
4. For those that live in manufactured housing, try to find a manufacturer that offers network cabling of any kind for rooms. (Despite the ease that this could be done in this housing market.)
5. Many consumers don't need 10 Gb in their home, what they need is something better than 6 Mb DSL (if this is even available) running from their home to the outside world. (And, lets not even mention data caps and the costs of going over said caps.)

I would love to have 10 Gb in my "tech enthusiast, double-wide manufactured home" (that I inherited from a family member) running over my ad-hoc mix of Cat5E and Cat6 cabling between my cable modem (60 Mb max, no data caps), 4 port 1 Gb router (self purchased), and "one per room" 1 Gb 4 port switches. However, is it needed? No, due to my recent move from DSL to quality cable (which was just recently expanded).

Wireless devices are, unfortunately, taking over. (Tablets, smartphones, etc.) In a decade or two, the traditional laptop computer may be obsolete. And, except for enthusiasts, the traditional desktop may be going the same way...

More and more AAA games are written for consoles, first now. In the future, as long as you have a "tablet" with the same specs as the latest generation console, and that tablet has wireless (lets say) 802.11ad or better...
 

Gillerer

Distinguished
Sep 23, 2013
360
81
18,940
The 2.5 - 10Gbps networks have very little to do with the speed of your connection to the Internet - 1Gbps is plenty fast for any external connection, if that's the only use case.

Instead multigigabit is used to connect devices in your home (or office) to each other.



An array of multiple fast hard drives can reach pretty high speeds for sequential workloads.

6 drives in RAID-Z2 without compression reached 488/429 MB/s (3904/3432 Mbit/s) (read/write) in a test here.

If the data is compressible (and compression is enabled - which is always recommended - the overhead is negligible), the read speeds will be much higher for ZFS: 7 drives in RAID-Z3 reached 1532/507 MB/s (12256/4056 Mbit/s) (read/write) (423/393 MB/s = 3384/3144 Mbit/s without compression).
 

drajitsh

Distinguished
Sep 3, 2016
131
20
18,695
@chrisramseyer
Please explain the following lines. In the 1 st it appears that you are saying that sfp is MUCH CHEAPER than RJ 45 while in the next line you appear to be saying the EXACT OPPOSITE.

SFP+ 10GbE switches are more cost effective. If you've shopped for a used 10GbE switch then you already understand the premium involved with the technology over standard RJ45 connectors.
 

Vatharian

Distinguished
May 22, 2009
90
0
18,630


If you have Cat 5 wiring spread around the house, that's not a studio flat, it's usually longer than 'shorter distances'. I've experienced signal dropping to 100M over 40-60 meters on older Cat 5 cabling, that's degrading after ~15 years or so. The external shielding and isolation often just crumbles away.

While renovating home, it's super easy to get Cat 6 cables, which, honestly, cost almost the same as your standard 5, and guarantees 10G to operate. From what I can see at my workplace, there are perspectives to run 40G on it, so it's future-proof... but swapping cables is one thing. More complicated PHYs are more expensive, kinda defeating the purpose - we want 10G brought to the masses, right? Sometimes sacrifices need to be made. We got rid of coaxial network cables (at considerable cost, to remind you), but with enough balancing and sophisticated electronics, cable companies can run 1+ Gb streams on them... but that is so expensive, only some companies can afford hardware for that.

I'd love to see 10G ending dominant connectivity tech. 100 MB/s is just not fast enough, not in an era of moving workloads away from the machine that's closest to you. And 4k cat videos ;) But jokes aside, I moved to 10G in my home, and while it costed me an arm and a leg, I have no regrets. Finally I could build reliable backup solution for whole family, strip my gaming PC from tons of (loud) disks without any kind of penalty, and made the data available to me without powering the power-hungry monster.

I probably need to point, that I live in Europe, where internet speeds in cities and even not that big towns that are below 100 Mb/s are often considered a joke, and generally fast internet costs peanuts. At my place I have choice between 250/20, 400/50, 600/50 and 750/750 Mbit connections from different providers, all of them are <$30/month and no one heard here about data caps, so I definitely see uses for really fast local network.
 



First you gotta understand what SFP's are and how they relate to a networking infrastructure. When your building out your layer one you decide your media type between devices and that media needs to terminate somewhere, typically a switch. Different media types have different electronic requirements, which quickly gets complicated in even the smallest datacenter. Try to imagine a networking switch having "four RJ45 ports, four MTRG, four LC-MM, four LC-SM" and so forth. It would rapidly become unmanageable as you have so many different media types requiring different port types. So instead they developed a universal port type where the media driver is a separate device entirely. You sell the switch as a sixteen SFP port switch and then let the network engineer buy whatever SFP's they need and using those. So a switch that use's SFP's for it's interface will be cheaper because it's doesn't the physical media drivers installed for every port, instead the user is required to buy the ones they need. A sixteen port switch would require sixteen SFP's to reach max capacity.

Now lets look at actual networking switches, the ones that don't use SFP's are usually 1000TX copper because 1000TX is a very cheap and very common port type. It's basically just RJ-45 with a Cat-5e or Cat-6 cable going to an Ethernet port somewhere. The more expensive fiber and twin-ax ports are usually SFP enabled. And then we hit the 10000TX standard which slams right into the laws of physics. The electricity and signaling rate required to drive such a signal down a twisted cable is expensive and hot. The SFP's end up requiring more power then the SFP port standard is allowed to give and that creates a whole host of issues. One solution was to use an on board PHY similar to how 1000TX switch's operate, but putting it inside the switch doesn't get rid of the power and heat generated from driving such a hard signal on commodity technology. So a 10GbE switch with on board PHY would be more expensive then a 10GbE switch that relied on SFP, but then you wouldn't need to buy SFP's.

Thankfully there are several manufacturers working on newer 14nm SFP's that have a power requirement under the SFP limit and thus can operate to standard without issues. This would then precipitate development of PHY's that can drive many 10GbE ports at low power requirements.
 

InvalidError

Titan
Moderator

Coax was never a scalable option for LANs due to collisions - only one computer can transmit through the channel at any given time. That's why coax as a LAN standard died with 10Base-2 in favor of switched 10/100/1000Base-T/TX over UTP and there is nothing fancy modern electronics can do about that, which relegates coax to backup/work-around status along with powerline notworking.

You can still use coax for LAN today using MoCA adapters. Still won't fix performance collapse issues with multiple simultaneously active endpoints as the collision rate increases.

As for cat5(e) cables decaying over time, my 20+ years old cables are still perfectly fine.
 

bit_user

Polypheme
Ambassador

If you don't have a fileserver or NAS, or do any other form of significant file sharing between computers, then we're basically just talking about sharing internet access that probably isn't (much) faster than 100 Mbps anyway (in the USA).

So, you're right about that, but I think it's safe to say that almost none of these people are PC enthusiasts and therefore not relevant to this discussion.
 

bit_user

Polypheme
Ambassador
I've got to echo what has been said about 10G BaseT, regarding heat, noise, expense, etc.

For the time being, I think 2.5G and 5G are fine compromises for most of us. I'd happily run 2.5G, if I could get a passively-cooled, 8-port switch for about $150 or less. 5G is what I'd really like, but I'm not willing to sacrifice on heat & noise.

Currently, I'm running a pair of used SFP+ 10G cards with a direct connection between them. I'd like to add at least one more PC, but would prefer to put my $ towards a more scalable solution than just buying another pair of SFP+ cards and cable.
 

parkerthon

Distinguished
Jan 3, 2011
60
52
18,560
It's not like 10G BaseT standards are that obvious. I just went through this exercise attempting to invest ahead executing a a retrofit wiring of a home I just moved into. It's clearly an area still under exclusive domain of the enterprise with the fast interconnects that they require along their core. That's why everything is still SFP. Even trying to look for ethernet cabling.... which is it? CAT6a, Cat 6e, Cat 7? And finally you get into buying them and they call themselves 6a or 6e, but they vary the shielding. So I start looking at 6a with the foil shielding around the individual pairs... fun times terminating that! No thanks. I ended up going with stuff that was called 6e but I have no idea how well it'll run 10gbe if it ever even gets there. Seems to me we're at least 5 years off if not more.
 

popatim

Titan
Moderator
You failed to talk about MSI's low cost 10GBe products which is what I was expecting to read about in this article. Only saying they can emulate the PHY in software doesn't translate to 'low cost' when they don't (and won't due to low demand) even have a product to sell...

Hey you know what, I have a plan for perpetual unlimited energy that never breaks down, but due to only being able to sell it once I won't bother...

Yeah, this is how disappointing this article is.
 

InvalidError

Titan
Moderator

There is officially no such thing as cat6e, whatever you may have that was labeled 6e is either incorrectly labeled cat6 or cat6a. Under ISO/IEC naming, those would be "class E" and "class EA" respectively, which could be where the 'e' came from and means that your "cat6e" really is just plain cat6.
 

CRamseyer

Distinguished
Jan 25, 2015
425
10
18,795
SFP+ cables are expensive (new or used compared to twisted pair copper). The used switches are cheaper than twisted pair copper switches online. This is mainly due to the products decommissioned by data centers and the demand for twisted pair by DIY types.

Any talk about 100Mbit should go to Craiglist or something. If a person is using a DLink hub from 2004 then they are not in the 10GbE discussion to start with.

1GbE is fast enough? How fast is that new SSD in your rig? HDDs can push 250 MB/s now and a 4-bay RAID 5 NAS can achieve 500 MB/s. Both of your end points can hit 500 MB/s but they are connected via a 100ish MB/s link.

The link, the point in the middle that we call a switch, router, whatever you use, is the bottleneck and the only part I wanted to focus on in the editorial.

Cat6 (and variants) cost the same as Cat5 (and variants). Most new homes should be built with at least Cat6 but we are also talking about enthusiasts and professionals here. We don't mind running cables across the floor in an office as long as they are tucked behind the desk with the rest of the rat's nest of cables.

Instead of thinking about a single user broaden the picture to a family. 10GbE is expensive to implement and all of the end products we're talking about (motherboards and NAS) come with premium pricing with the 10GbE feature. Disposable income usually comes later in life as do families. A family of five has how many connected devices? Everyone in my home has at least two PCs.
 
Status
Not open for further replies.