Switches, Hubs, and Modems
cancel
Showing results for 
Search instead for 
Did you mean: 

Failover network with ProCurve 2626

SOLVED
Go to solution
Alessandro_78
Regular Advisor

Failover network with ProCurve 2626

Hi all,
actually we have a single cisco router (2800 with two ethernet) connected in a unmanaged switch with a bouch of servers.

We want to make a fully redundant network (except for the single router) not only at L1-L2

My plans are to connect each server (with two ethernet in failover) to two ProCurve 2626 (one ethernet per switch) and then connect each switch connected to a single ethernet port on the router.

Something like this:

CISCO 2600
/ \
/ \
HP 2626 HP 2626
\ /
\ /
BOUNCH OF SERVERS

Is it correct?
Should I connect together the two switches?
RSTP should be enough or I'll must implement other protocols?

Are there any single SPOF on the network side? (not counting the single router)

Sorry for my bad English but I'm Italian.
14 REPLIES
Evert Goor
Trusted Contributor

Re: Failover network with ProCurve 2626

Looks ok for me and make also a trunk between the 2 hp2626 switchs to have a better inter server connect.
There are some problems between in rtsp between the cisco's and the procurve switches. Make sure they are on the latest firmware level and test it well before accepting this in a production environment.
Evert Goor
Trusted Contributor

Re: Failover network with ProCurve 2626

Looks ok for me and make also a trunk between the 2 hp2626 switchs to have a better inter server connect.

There are some problems between in rtsp between the cisco's and the procurve switches. Make sure they are on the latest firmware level and test it well before accepting this in a production environment.
Alessandro_78
Regular Advisor

Re: Failover network with ProCurve 2626

Is RSTP mandatory or I can do the redudancy without it ?

On server (Linux) i'll make a failover with bonding (one ethernet active and one passive).

What if the active ethernet is eth0 and the active ethernet on the router is ethernet1?

Packets going out from ethernet1 on the router goes to the right switch and then on the eth1 on the server that is down.

With a trunk between switches packets will be redirected from the right switch to the left swith and then on the server's eth0?


Doing so, i'll archieve full redudancy and I'll can power down one switch or unplug one ethernet cable without loosing connectivity. Isn'it?


Again, sorry for my English.
Mohieddin Kharnoub
Honored Contributor

Re: Failover network with ProCurve 2626

Hi

You can do the following:

1- Create a Trunk between the HP Switches with 2 or more links.

2- Tag all the Vlans on that trunk between the switches.

3- Now if you connect each Server to one of the 2600 switches, you can do Teaming.
Teaming is like creating one Virtual interface on the Server combining the 2 Network Cards, with one IP address assigned.

Usually there will be a utility for that oin your Server.

By doing this, you will be sure that you have achieved: Redundancy and Load Balancning

So both Network cards in your server will be active and forwarding traffic by Teaming them.

On the Cisco router, you can do something called Grouping to multiple interfaces into one virtual one for redundancy and load balancing purpose.

But i'm not sure on which IOS version you can get this feature, but its better than using spanning tree.

Other wise Just enable the Spanning tree, but by using RSTP you will have one link down from one 2626 switch to the router.

Good Luck !!!
Science for Everyone
Evert Goor
Trusted Contributor

Re: Failover network with ProCurve 2626

can you bond the ports in the cisco to one interface/vlan then it can work but all traffic between the 2 hp2626 switches will go over the cisco. It can be done but it is not the nicest solution. rstp is need to prevent unwanted loops in the configuration. I prefer to have it allways working.
Alessandro_78
Regular Advisor

Re: Failover network with ProCurve 2626

To Mohieddin Kharnoub:

1) Ok for the bonding
2) Ok for trunking. I prefer trunking than RSTP but in normal situation (both switches operative) traffic going from one server to the router, should pass on the trunk or goes to router directly?
I'm asking so because i have only one gigabit port available for trunking and i cant aggregate more...

Actually i don't use vlans, can I do redundancy in the same way?

Thanks.
Mohieddin Kharnoub
Honored Contributor

Re: Failover network with ProCurve 2626

Hi

First of all, trunk is Two or More physical links aggreagetd together to form one logical Link.

Since you have ONLY one physical port, then its called an UPLINK.

Now, if you do Teaming for both network cards in the server, and from the router also
I believe the traffic will flow over Both links from Server to Switches, and from Both switches to the Cisco router.

Just think about the 2 Bounded interfaces in the Cisco router are ONE.

And from the 2 teamed network cards on the Server are ONE.

So i think this is Load balancing + Redundancy.

Good Luck !!!
Science for Everyone
Alessandro_78
Regular Advisor

Re: Failover network with ProCurve 2626

Ok for trunking.
But, the uplink between switches is needed to have more redundancy? Like my example:
if eth1 on the cisco fails, all traffic is redirected on the other switch with the uplink?

Is correct? Without uplink, if one ethernet fails some packets may be lost (going out from on ethernet to a switch that isn't connected to the server's ethernet because link failed). With uplink switches routes the packetes on the other switch.

Without vlan, can I do the same?
Why there aren't path-loops? Uplink should make loops...
Mohieddin Kharnoub
Honored Contributor

Re: Failover network with ProCurve 2626

Hi

Thats true

Your uplink needs to be A Trunk (two or more physical uplinks) between the switches.

Now for the Loops, if you didn;t group both ethernet interfaces on the Cisco router and you have logically 2 interfaces, then

yes there is a loop.

Thats why i recommended you to bind them into one logical interface.

My Recommendation:

- Free up 2 interfaces on each switch (at least) and create a Trunk.
- Group both Cisco router interfaces.

Vlans, is not an issue here.
If you have Vlans and you need to enable routing between them, then you can do it on the 2600 switch or on the Router.

Don;t forget to assign points to the posts that helped you.

Good Luck !!!
Science for Everyone
Mohieddin Kharnoub
Honored Contributor
Solution

Re: Failover network with ProCurve 2626

Hi

Thats true

Your uplink needs to be A Trunk (two or more physical uplinks) between the switches.

Now for the Loops, if you didn;t group both ethernet interfaces on the Cisco router and you have logically 2 interfaces, then

yes there is a loop.

Thats why i recommended you to bind them into one logical interface.

My Recommendation:

- Free up 2 interfaces on each switch (at least) and create a Trunk.
- Group both Cisco router interfaces.

Vlans, is not an issue here.
If you have Vlans and you need to enable routing between them, then you can do it on the 2600 switch or on the Router.

Don;t forget to assign points to the posts that helped you.

Good Luck !!!
Science for Everyone
Alessandro_78
Regular Advisor

Re: Failover network with ProCurve 2626

Ok for the loops.
But, why should i trunk more than one link between switches? Using one link only what's the problem.?

I checked better:
actually I have 20 port unavailable switch (20 server),

I'll do the following:

1) group ethernets on cisco
2) bonding ethernet on server (load-balancing mode and not active/standby)
3) trunking the 2 gigabit port between switches (so I'll have one 2000mbits logical link).
4) connect the port 24 of each switch to one ethernet on the router.

5) optionally, I can add port 21,22,23 to the trunk, so I'll have 2300mbit trunking.

The trunk will be useful in case of ethernet failure.

Thanks for the reply.

You'll got a 10!
Alessandro_78
Regular Advisor

Re: Failover network with ProCurve 2626

Hi,
I read something around and i've seen that is not possibile to have load-balancing with two switches like mine.

I can make a trunk between switch (LACP or trunk? which is better? I think lacp.) but i must make an active/standby bonding on servers and router because to have load-balancing (active/active) the switches's ports must be in trunk and I can't make it trunked because they are on different switch.

So i'll do the following:

1) group ethernet on cisco (active/backup)
2) trunk 2 gigabit ports on switches (may be LACP active/dynamic)
3) bonding two ethernet on servers (active/backup)

Thanks a lot.

hagrno
Occasional Visitor

Re: Failover network with ProCurve 2626

No, you can not use 26-series switches for active redundancy. You'd need a 35 or 54-series switch with a Premium License supporting VRRP (virtual router redundancy protocol), but then again you'd need to use them as routers.

Or, you could use mesh-switching which would make two switches one logical switch unit. This is not supported on the edge products, though...

The 26-series can at best do failovers with STP as someone already posted above. Always use the latest STP if you're able to, as STP(1) is slow to turn the network map around.

On the servers you'd want one NIC on each switch running in active/passive. You have several ways of doing bonding (trunking) on them, but you'd lose more IO than what you're spending if you want a full trunk since the 26-switches can't trunk interports.

Hope this helps
Alessandro_78
Regular Advisor

Re: Failover network with ProCurve 2626

Ok for the active/active.
I want to make a redudancy network with active/passive without using STP.

Each switch will be connected to 18-20 server, so in case of failure of one switch, all traffic should be redirected to the other via trunk cable.

I'll trunk switches with 2 gigabit port on each, so I'll have a 2gbits trunk.

In case of failure and in case of heavy-loaded servers, I'll have at maximum 20*100mbit of bandwidth: 2gbit, the same as the trunk.

So, is not correct do the following?

- configure cisco with active/standby ethernet
- connect switches with dual gigabit ports
- connect each server to each switch
- bonding each server with active/passive.


Sorry for English.