upgrading-lan

Hello,

I need some quick advice on upgrading our lan as I am a bit unsure on a couple of things. We are intending to buy netgear 4 24 10/100 switches and 2 24 gbit port switches to replace our old ones. The switches I believe all have stack ports and 2 2gbit uplink ports and was told I can stack up to 4 of them.

I have a couple of questions

1) Someone was suggesting not to use the stack ports but just connect the gbit ports from the 4 10/100 switches to the gbit switches. I am a bit wary of this and I think this could cause bottleneck - I am correct? If 10 machines were trying to talk thorugh the gbit port to the other switches then wouldnt that saturate the link? - whereas if you used the stackable ports to interconnect the switches this would not happen?

2) If you do 'stack the switches' then isnt there a redundancy problem in that in effect the stack ports are daisy chained so if the middle switch broke this would affect communication between switch 1 and switch 4? So for redundancy if you elimetaed the stack method but connecting each switch to the gbit backplane switch, if one of the

10/100 switches died - it would not affect any of the others?

Thanks for any advice.

Reply to
chris-c
Loading thread data ...

In article , chris-c wrote: :I need some quick advice on upgrading our lan as I am a bit unsure on :a couple of things. We are intending to buy netgear 4 24 10/100 :switches and 2 24 gbit port switches to replace our old ones.

If you intend these to be high utilization devices, then it would be a good idea to buy 1 first and stress test it in your environment. Different Netgear switch models have different quality levels. The further you get from Netgear's traditional consumer-level device, the more caution is advised.

:1) Someone was suggesting not to use the stack ports but just connect :the gbit ports from the 4 10/100 switches to the gbit switches. I am a :bit wary of this and I think this could cause bottleneck - I am :correct? If 10 machines were trying to talk thorugh the gbit port to :the other switches then wouldnt that saturate the link? - whereas if :you used the stackable ports to interconnect the switches this would :not happen?

Switches that are directly stacked will use the local stacking bus to exchange data. The speed of that will depend upon the speed of the stacking bus and upon how it is used [some stacks are sychronous, alloting time for each stacked device in turn, even if they have nothing to say!].

If you have two switches without uplink ports that could stack together but you connect them by way of a different switch, then the latency will be the latency of serializing the data out of switch 1, receiving it and switching it in the core switch, serializing it out of the core switch, and receiving it in the destination switch; the maximum throughput for this operation will be the minimum of speed of the links involved. In the case of the 24-port 10/100 switches, that's going to be a maximum throughput rate of 100 Mbit/s, with latency of two switches serializing for transmission and two switches receiving transmission and one switch switching between ports; add a tinsy bit for the actual wire speed between the devices.

If you stack the two switches together instead, then the latency would be the latency of serializing to the parallel shared bus, and of receiving at the other end of the parallel shared bus, plus an itsy bitsy teeny weeny time for transmission over the 1 metre bus cable. Throughput would be whatever the rate the bus ran at, taking into account parallel transmission rather than serial (e.g., 100 MHz of parallel 8-bit data is equivilent to 800 MHz of bit-at-a-time serial data.)

Stacking busses are usually noticably faster than the fastest port. For example, on the 24 port 10/1000 switch, the stacking bus might run at 1 Gbit/s.

:2) If you do 'stack the switches' then isnt there a redundancy problem :in that in effect the stack ports are daisy chained so if the middle :switch broke this would affect communication between switch 1 and :switch 4? So for redundancy if you elimetaed the stack method but :connecting each switch to the gbit backplane switch, if one of the :10/100 switches died - it would not affect any of the others?

It depends on how the manufacturer impliments stacking. Nortel and Cisco use ring topologies (last device is connected to the first) so there is always a way to get to a device if something in the middle breaks. The better switches automatically manage around failures. For example, Cisco's new "StackWise Technology" is fully able to have a new switch step in quickly as the new master if the master switch fails.

In the star topology you haven't eliminated single points of failure: you've just made the gigabit switch your single point of failure. Considering that gigabit is newer technology that hasn't had as much time to smoke out the bugs, you could be -reducing- your reliability by going star topology instead of stacking well-developed 10/100 switches.

Reply to
Walter Roberson

nortel stacking used to support more than 4 devices so i am not sure if this is correct - but you dont specify which switches.

also, nortel tend to have several models that can be in the same stack.

the golden rule is to work with accurate info. i suggest you look at the web site and work out what the boxes can do:

formatting link

depends on the aggregate traffic. If your network only has 1 or 2 main servers connected at Gigabit speeds, then the servers will form the practical limit to throughput rather than the switch ports no matter how you interconnect the switches.

if the switches have built in stacking, then stack cables are probably cheaper than GBICs (assuming they dont come with the switches as standard anyway).

Stacking has always been designed to be faster than Gigabit - high capacity between the switches is one of the major selling points.

nortel stacks usually support a "ring" of stack cables - the idea is you can loose a cable or a switch, and the rest of the stack isnt affected.

the main drawback to stacking is that the switches in the stack end up very closely linked, so there may be sortware version constraints, and the risk of a software bug affecting the stack is probably greater than if you just linked devices together via Gigabit - the switches in the stack are logically more like blades in a chassis than standalone boxes.

having said that nortel have been stacking switches for a long time, and it has worked well for me.

Reply to
stephen

Definitely connect the gig ports on the 10/100 switches to the gig switch directly. You'll get much better overall bandwidth - think about it - if you stack them and then run from just one gig port, you only have 1/4 the bandwidth going to the gig switch.

Yes. The way netgear stacks, yes. Other switches can stack in a ring - top and bottom of stack connect - so they are tolerant of a link failure in the stack. But not netgear.

MY biggest suggestion, after sturggling to get netgear switches to work as advertised over the last year and a half, is AVOID NETGEAR HIGH END SWITCHES AT ALL COST.

I have had VERY bad luck with the following:

GSM712, FSM726s - bad handling of multicats packets - very easy to lock up a port with multicast traffic.

GSM7312-GSM7324 - The ospf implementaion is so bug ridden as to be almost completely unusable.

Reply to
T. Sean Weintz

Thanks for all the help here.

We looked at 3COM stacking solution but I find with 3COM switches you need to buy every single module e.g. stack kits, gb cards, which is not so straight forward.

We had quite a good solution from HP which looked as follows 2 pairs of switches connected to 2 GB switches. The servers would be on the GB SWITCH1 and the clients on the 10/100 switch.

switcha(10/100) GB1 GB2 | | Trunked 2GB to GB switch1 --2GB---> GB SWITCH1 switchb(10/100) GB1 GB2 |-----------^ | | |8GB TRUNK switchc(10/100) GB1 GB2 | | | | Trunked 2GB to GB switch1 GB SWITCH2 switchd(10/100) GB1 GB2

The GB SWITCH2 would act as a cold failover.

Has anyone had any experience with HP switches? ie any good/bad references?

Thanks

Reply to
chris-c

As you can see from above, your ascii diagram got clobbered upon posting, or at least I THINK it did -- either way I cannot make sense of it.

Whay HP model switches are you looking at? The one thing I can about switches is if you buy cheap stuff, you always get what you pay for (ie: cheaply made equipment)

I have heard good things about some of the HP switches. I have also heard mostly only good about Extreme. I have used Nortel quite a lt and had good luck with them.

But you can't really do it by brand - you need to look at each specific model seperately.

Reply to
T. Sean Weintz

The current series are not as outstandanding as the 4000M series where back then but we still buy them for their lifetime warranty. Personally, I try to avoid stacks in one rack for management, performance and reliability reasons. Even if it is not the cheapest solution, I would go for a chassis switch. With HP I would buy the bundles, put the modules and the power supplies in one chassis and even have a cold failover chassis.

Reply to
Manfred Kwiatkowski

Cabling-Design.com Forums website is not affiliated with any of the manufacturers or service providers discussed here. All logos and trade names are the property of their respective owners.