Core Switch issues

[sluggster66]

Hi everyone

 

Looking for an explanation to do with ARP tables.

 

We have recently had a new core switch installed and our network Vlanned.

 

We are having an issue with client machines getting a message saying 'there are no logon servers to service........'

 

This is happening to different clients on different vlans at different time. What I have noticed is that during quiet times on the network we are not getting this issue.

We have had the company that installed the core switch and Vlanned the Network out to look at the issue, and it appears that the core switch has a limited arp table of 256.

 

 

What does this mean?

 

And would this be the expected effect of going over this limit

 

Many thanks

[synaesthesia]

256 addresses? what's your core, a 4 port netgear home router?

 

To start with, as 512 is pretty much the minimum I see on the cheap and cheerful (see £100 24port gigabit switches), that seems to be either false or your core is something really old/not designed to be a core switch. Basically it's the way switches convert IP addresses to MAC addresses and vice versa. So if on a flat or non routed network you have more than 256 networked items (other switches, printers, PCs, anything connected simultaneously) you'll hit that limit quickly. Easily solvable on a correctly set up network but without any more information I'm not sure where you'd start.

[m25man]

Whilst the described issue is not uncommon - under specified core switch hardware in relation to number of hosts on routed network, it can also be caused by a single mis configured host causing ARP poisoning.

Watch out for simple mistakes such as incorrectly set subnet masks on manually configured devices, this can set the broadcast address out of band for that device which can cause havoc in a small LAN with limited switch capacity and lower powered CPUs etc.

I have seen very slick expensive switchgear reduced to less than C64 performance through simple host configuration errors.

 

How many hosts on your entire network? What is your Core switch choice?

[twin--turbo]

256, that's got to be wrong.

 

Rob

[m25man]

Yes, I can't think of a L3 switch with less than an 8192 ARP cache and then if the VLANs have been set up correctly any broadcast issues will be restricted.

I suspect a config error somewhere.

[ChrisH]

Have the machines been on a while or are they doing this freshly booted ?

[sluggster66]

Thanks for the replies.

 

We have about 650 devices hanging off the network.

 

The core switch is a new HP 48 port gig switch, not sure of the model without checking.

 

The problem we have had is that the network runs fine until the network gets busy with the majority of the devices coming on, then we have machines bring up the ' there are no logon servers to service ' message come up. As the network gets less busy ( end of school day) the problem machines then work.

 

The.arp table with 256 is something I was told.

 

Thanks Simon

[sluggster66]

Have the machines been on a while or are they doing this freshly booted ?

 

 

Some have Been on and working other have just been turned on.

[ChrisH]

If they were just the ones turned on I would be looking at spanning tree settings on your switches and if they are not going to forward quick enough and delaying DHCP, but since it's happening to machines that have been on a while I would be looking for something causing a broadcast storm maybe. This is where a system like HPs PCM is invaluable else you will need to be checking the management pages and logs of each switch.

[DMcCoy]

Thanks for the replies.

 

We have about 650 devices hanging off the network.

 

The core switch is a new HP 48 port gig switch, not sure of the model without checking.

 

The problem we have had is that the network runs fine until the network gets busy with the majority of the devices coming on, then we have machines bring up the ' there are no logon servers to service ' message come up. As the network gets less busy ( end of school day) the problem machines then work.

 

The.arp table with 256 is something I was told.

 

Thanks Simon

 

The arp cache limit on the v1910s (rebranded 3com) is 256 entries. I've not found any specified limits for the lowest "E" procurve that supports VLAN static routing yet.

 

What is the switch? Although I'm guessing 1910 now!

 

EDIT: found the table size for the 2610 (lowest current model with routing, 8000 entries). Lets all hope it isn't a 4100, I've liked all the procurves since, but the 4100s were terribly underpowered and only really suitable for L2!)

Edited September 7, 2013 by DMcCoy

[seawolf]

Thanks for the replies.

 

We have about 650 devices hanging off the network.

 

The core switch is a new HP 48 port gig switch, not sure of the model without checking.

 

The problem we have had is that the network runs fine until the network gets busy with the majority of the devices coming on, then we have machines bring up the ' there are no logon servers to service ' message come up. As the network gets less busy ( end of school day) the problem machines then work.

 

The.arp table with 256 is something I was told.

 

Thanks Simon

 

The answer to your problems is to hire someone who knows what they are doing to come set up your network. It's obvious that whoever you've paid to set up what you have now doesn't know what they are doing. So, get on the horn to an actual HP ProCurve network engineer in your area and your problems should be solved in a very short time and you'll have a network that performs very well.

Edited September 7, 2013 by seawolf

[seawolf]

The arp cache limit on the v1910s (rebranded 3com) is 256 entries. I've not found any specified limits for the lowest "E" procurve that supports VLAN static routing yet.

 

What is the switch? Although I'm guessing 1910 now!

 

If someone has sold them a 1910 as a core switch they definitely need to be fired...after asking for a refund.

[synaesthesia]

1910's are fine as edge switches, we use them with 600 clients without issues but when a single switch is expected to handle all core<>edge comms as well as directly connected clients. I use them as "core" switches when there's 50-60 stations (primaries) without a problem but really is dodgy ground otherwise!

[DMcCoy]

Ok, even the wheezy 4100s had an 8000 entry routing table

[seawolf]

1910's are fine as edge switches, we use them with 600 clients without issues but when a single switch is expected to handle all core<>edge comms as well as directly connected clients. I use them as "core" switches when there's 50-60 stations (primaries) without a problem but really is dodgy ground otherwise!

 

1910s are fine for an edge switch (I use one for that purpose), but as a core switch for 650 clients - what the...

[DMcCoy]

As for the issue:

 

MAC address table is needed to know which devices are on which ports

ARP table maps IPs to MAC addresses and is needed to know which IPs are on which VLAN/port.

If you run out, then you will has issues as you have found, it may not learn any new MAC/ip locations until it removes or expires an old one

 

You may be able to help slightly with a shorter arp timeout before it's removed, but this is not a good solution.

 

If you had been *just* over then you may get away with 120 second timeout or so, but over twice the number of machines as entries is going to be a problem. At least if it is a 1910, it was cheap.

Edited September 7, 2013 by DMcCoy

[twin--turbo]

As for the issue:

 

MAC address table is needed to know which devices are on which ports

ARP table maps IPs to MAC addresses and is needed to know which IPs are on which VLAN/port.

If you run out, then you will has issues as you have found, it may not learn any new MAC/ip locations until it removes or expires an old one

 

You may be able to help slightly with a shorter arp timeout before it's removed, but this is not a good solution.

 

If you had been *just* over then you may get away with 120 second timeout or so, but over twice the number of machines as entries is going to be a problem. At least if it is a 1910, it was cheap.

 

It will learn a new address almost instantly, but that requires a broadcast. As every clinet sends a request any destination not in the ARP table will cause a broadcast. As this continues the switch will get bogged down sending, reciving and processsing arp entries and will grind to a halt very quickly. It's CPU will spend time on constant persistance ARP work and not data transfer.

[sluggster66]

Update:

 

The switch has been changed and everything working again.

 

Thanks for all the help.