Ethernet switch?

As is usual with these things, if it ain't broke, don't fix it!

Do you have any issues?

As long as everything is working fast and reliably, I wouldn't bother unless you run out of ports on the AE. You'll have another little box cluttering the place up, another socket with a pesky SMPS in it, and one more Ethernet cable to deal with.

i use a switch to connect my streamer and server plus an Airport Express (but this only has one port.) The reason I have this arrangement is that my router/phone line are a long way away, so I use a wireless connection for iRadio etc, but still keep all the local streaming wired.

Depends how these are connected. Best practice dictates that a switch between two end points is always recommended because it facilitates reliable flow control - even in a network comprised of only 2 hosts.

But as far as I know, the Airport is also a switch and an access point (maybe I am wrong on this).

Although I always go for best practices, if you have been connecting the UQ in the same way so far with no problem there is no point in messing with it. i.e. if the connection is currently stable then there is no problem which adding a switch would resolve.

Really ? I think not..... You would only use a switch between two hosts as a last resort if you didn't have a swap over cable... also a consumer (layer2) switch does nothing for flow control... Flow control is a transport layer function which is layer 3, a switch is oblivious to this.

A switch can prioritise frames, but between two hosts there is no priority !!

Simon

If you search the forum, you'll find lots of discussions on this. Some people claim they hear an improvement with Cinnamon or other cables, others believe such a thing is impossible. I'd suggest looking for a dealer who will let you borrow one for audition, compare it to regular CAT5e or CAT6, and see for yourself.

Yes, I wouldn't get hung up with Cat6, Cat6a, Cat 5e. The attribute that will most likely effect the radiated and conducted EM fields ( this is typically what causes the Ethernet patch lead to 'sound' different on connected equipment) is whether they are not shielded (UTP), shielded (STP) or foil shielded (FTP). Both Cat 6 and 5e come in UTP and STP variants, check the sheath on the cable.

Simon

eeeekk   its the ethernet cable question again.  

Try searching this forum,  loads of posts & endless discussions

One camp says there is no difference & support that with various technical papers, test data & blind listening tests.

The other camp says there is a difference,  problem is its all subjective "trust me I'm an expert" etc.

As for myself,  I've heard a very small difference when trying AQ Cinnamon (plus other Cat6 & 7's) against a freebee Cat5 that came with a BT wireless hub.  It was a well used & I suspect it might have been damaged & the twisted pairs out of geometry.

But comparing AQ Cinnamon to Supra Cat7+ & a good quality brand new Cat6,  I heard no difference whatsoever.    

Sorry Simon but you are wrong on this. Your bog standard L2 switch does contain a buffer which can help greatly in two endpoints managing their own flow control in a resource effective manner. And as for best practice, it has nearly always been the case that appliance vendors do not recommend direct connectivity even for the most basic of networks. In fact many vendors I have worked with like IBM, HP and Safenet won't even support connectivity issues between their hardware unless the topology includes a switch.

Ok we will quite emphatically have to agree to disagree on this one . absolutely not the case in my world, my customers and my design patterns globally.

I can't help feeling suggesting such things in the special back to back two host case would fail you many vendor accreditations. Unless i am mis understanding what you are trying to say I just can't technically possibly think what you are talking about with regard to a two host communication.. In such a set up the only thing I can see a layer 2 switch could do would be to drop corrupt frames, but that does not affect flow with two hosts.

Multiple hosts can benefit from  802.1p prioritisation for QoS but that is prioritising multiple port queues on the switch to a single port.. Ie merging traffic in a prioritised way... But here we are switching multiple collision domains which is exactly is what a switch is designed to do. But the special case of back to back connections between just two hosts we use a purpose built cross over cable to effectively create a single collision domain between hosts providing duplex transmission.  

For example such a topology is put in place on back to back router setups between vendor VPNs.. as a router in its pure sense is a host. Adding a switch here would add a point of failure for no advantage and therefore would be / is  discouraged as one of my CCIE engineers has previously reminded me on exactly such a topology.

Anyway off to work now, ensure these global and corporate networks me and my teams have been / are responsible for designing are still all up and running smoothly...or perhaps I will have world war three in my inbox with all these back to back networks failing.......

Simon

I think we are talking at cross purposes here Simon. I am not referring to QoS or VPN but fundamental connectivity here.

There have been a number of setup guidelines I've ingested over the years for server appliances from vendors that specify a switch even in the case of a 2 node network using the language "best practice". And as mentioned, a baseline for their support of connectivity issues was often "put in a switch and then we'll go from there".

Don't get me wrong, I'm not in any way indicating I think this is a technical requirement. It isn't. If both nodes are functioning, RFC2239/3635 compliant and setup correctly, a L2 switch has nothing to offer other than the buffering (which is minimal in domestic use switches - often lower than 128KiB for a 5 port switch which likely only supports auto negotiation with a values less than the max supported by the endpoint NICs).

Do I set up simple 2 node networks directly connected? Yes. For lab testing. When I am confident in the operation of the NICs and I control their configuration and more importantly, the support buck stops at me. Incidentally, any appliance I have directly attached to another built in the past 15 years has never needed a Xover cable.

Do I mean a switch between just any endpoints? No. You are quite right, you wouldn't use one without good reason between VPN boxes (though it would be mandatory to use a switch for a resilient VPN box cluster for failover to function reliably in all scenarios). Similarly, it may not be required between a number of network appliances where the inclusion would simply be adding a box for the sake of it.

Have I seen instances where a basic L2 switch resolves connectivity issues? Over the hundreds of issues that have crossed my desk over the years, only 2. In both cases one of the end points (a Juniper VPN box of the same model) had a firmware bug that prevented auto negotiation occurring in an RFC2239/3635 compliant manner leading to high numbers of packet collisions). But having supported a single managed VPN for 10 years with several thousand boxes in the field over 200 countries, the fact remains that when partners or vendors are all stipulating the same "best practice" and you don't follow it, come audit time there are things to answer for. Unfortunately I work in the real world where your best judgement aside, adherence to practices and accountability become mandatory considerations of doing business and the same policies enforced by your partners and vendors need to be extended and enforced to your customers as a prerequisite of doing business and getting people around the table when problems happen. From the technical perspective, on that scale it make complete sense to ensure everyone is doing things in more of less the same way otherwise it becomes unsupportable.

It is also pretty rare for a crossover cable to be required these days even for direct connected VPN boxes, such is the current ability of ports to detect this and adjust. I used to keep a few crossover cables around the data centre but after the last few ancient Cisco devices were disposed of, they were never required again. Mainly kept around for legacy stuff. 

A decent L2 switch with a reasonable frame buffer (1GiB or more) will allow two endpoints to independently negotiate their highest speed and duplex irrespective of the other node and, within reason, accommodate for that differential in speed to the extent that the buffer can cope with it.There are sadly also a number of NICs out there that are not RFC2239/3635 compliant. A sophisticated switch can also detect and adjust for larger differences such as hugely mismatched MTUs where one node is spitting out jumbo frames at MTU9000 and the other only supports MTU1500 by actually breaking apart the larger packets into smaller ones within the supported limit. Otherwise, the two hosts, directly connected, would barely talk to each other. 

Zen out.

Both Simon and Zen are clearly much more expert on these networking issues than I am, but my own personal experience is that if you connect things via an unmanaged network switch, the network discovery works more reliably than otherwise.

I found that using the "switch" in my D-Link router often gave me problems where my Thinkpad Windows 7 PC couldn't see the downloads folder (or any folders) in my Unitiserve, although an old HP PC running Vista could do so and similarly my Thinkpad couldn't always see my Brother networked laser printer while my old PC could.  Sometimes re-booting the router would resolve things and sometimes re-booting the Thinkpad would, but other times rebooting neither would.

Now I have everything connected into a network with Netgear switches in several places around my house all interconnected and a single Ethernet connection back to the router. It all now works all the time and always reliably. I have four Naim streamers which can always see my Unitiserve and my IPhone/iPad any time I want.

This is the origin of my advice I give often in this forum that connection of Naim streamers, servers, NAS etc should be via a cheap switch like the Netgear GS105 rather than the switch inside the router.  I regard it as good housekeeping which helps to avoid problems caused by poorly conforming consumer networking components not quite working together properly all the time.

best

David

+1 x 100%  ...........  Last Sunday I spend a few hours trying to fix a friends intermittent Linn-Qnap hooked up via an ISP supplied latest & greatest SuperHub.  I couldn't sort it & to me it was a problem with the router;  to prove it I nipped off home to borrow my GS105 & a few lengths of well bashed Cat5  -----  "BINGO"

He was surprised at the new connection response time & is pretty convinced it sounded better.

He has a shopping list for a switch, some ethernet patch leads & new wall jacks,   & hopefully I can complete on Saturday.

+2.

Could this be the reason my woefully underspecified TS112 NAS and SBT work together at blistering speed.

Interesting thread, I can't get my hear round how two people who obviously know their onions (and more) when it comes to networks, have completely opposite views.

Phew, just got back, world war three hadn't broken out... My back to back cross overs are still working around the world and all fine and flowing well at 1Gbps and 100Gbps. I did check my old Cisco guides , and back to back connections on a 2 host network, use a crossover network cable.. With wiring explained.. And that appeared in the ICND1 book, ie the most basic foundation level guide.... Anyway moving on you ... 

Fatcat, there are usually many ways of achieving an outcome..my design principles are to remove un necessary complexity and redundancy thereby increasing reliability... feeling zen had specific reasons in adding thsee additional components in his designs which became clearer to me in his  later posts, without those specific reasons then it is unnecessary on a two host connection and otherwise would reduce reliability.

Now ow regarding the OP and Mikes response.. some consumer router switches contain hardware switching chipsets as opposed to software chipsets. If hardware switching is incorporated in a router switch, then there is almost certainly no advantage in using a seperate switch other than physical layout.. The issue is however it is often hard to understand how the consumer router switch is designed and therefore a seperate switch is a safety net. I will poke around the web to see if I can find out how the Apple devices implement thier switching capability.

Simon

In my case last weekend, the hub had worked OK for about 1 year as an audio streamer switch.  Something had gone belly up.   That adds to my growing mistrust of IPS routers in that they may or may not incorporate a genuine switch functional - they never tell us - but I very sure they contain minimal build quality & reliability.

Ok the latest AirportExtreme uses a Broadcom 2 core CPU with integrated hardware USB and gigabit switch chipset, namely the Broadcom BCM53019A1.

This device appears to offer hardware accelerated switching and so I think it unlikely there would be any performance issue between switch ports ( ie the switching function) but one can't tell for sure .. but performance between switch plane and wifi and switch plane and wan router could be more easily limited by resource limitations

So you still think the AE is all I need or do you now recommend a switch? TIA

I kind of feel you should be alright.. so if everything is working ok, I think its unlikely you are missing out  on anything.

Simon

I'm going to duck!

(I know the principles, but not the details to this level!)

A member of my family who is a professor teacher has said to me " explain it to me as if I am a complete idiot "  this always works for him.

This is excellent advice and easy for me because I actually am an idiot.

Refer to the second post at the top of this thread by garyi - excellent advice worth heeding