IRF 2 on A7500

[daperzoid]

Hi, I have the following doubt: If I form an IRF2 group with my 2 A7500 switches, each with dual engines, will I still have 48Gbps per slot?.

Cheers,

Dan

[Fredrik Lönnman]

Hi,

Im pretty sure that both fabrics will be active on both your switches in the IRF, though I cannot find any reference at this point.

Regards,

Fredrik

[daperzoid]

Thanks for your answer mate. So, with this implementation do we have a stateful failover?. Since sessions are splitted between all 4 fabrics, if one fabric all session that were using that fabric will have to be restablished right?. And how will load sharing between both chassis work in this situation, were one chassis will have only half of the switching capacity?. These questions are very important for data center deployments and customers are concerned about it.

Cheers,

Dan

[manuel.bitzi]

Hi daperzoid

May I can help you a bit.

The usual traffic dont goes through the fabrics all the time. the fabrics i.e. calculates routing and switching tables and writes it to the blades. then the blades can decide be themself where to route or switch the traffic to. i.e. they can send the traffic directly from one port to an other without contacting the fabric, they usally search the fastest way.

only one fabric calculates. the others are standby and sync the calculated datas with the master. if the master fails, a new master is elected and continous the calculation.

all chasis and blades will habe there full switching capacity

If you habe more questions, please ask more specific. So I can give you a specific answer. :-)

br

Manuel

[SINDIBADE]

I have a question concerning the same subject. I will implement two switches A7506 using 2 switch fabrics through a 10Gbps ports on each. can they be used to form an IRF stack to aggregate the capacity of the four switch fabric?
can we distribuet link aggregation DLA between IRF stack with another XRN stack using E5500 GE?

regards

[manuel.bitzi]

@Sindibade

Yes to both questions.

2 Links between the IRF-Members are aggregated and recommanded => if one links is broken, there is no split brain.

DLA is a IRF interna lfeatures only. Für the E5500G, it's a normal LA-Link.

br

Manuel

[sindibade_1]

i assume that i can do LA on both 10GB port switch fabric on both switchs that can form 20Gbps IRF LA?

[Fredrik Lönnman]

Yes you can.

[sindibade_1]

thank you so much.

[sindibade_1]

Hello,

I have successfuly complete the basic IRF configuration on 2x A7506 switchs and i have succefuly create the virtual switch from both.

I need to know what is the best scenario to implete the MAD protection mecanisme suching that i will use a LA to connect all access switchs.

Is LACP MAD is necessary for this context?

IRF link is formed from 2 grouped TenGig XFP ports on both switchs.

best regards

[manuel.bitzi]

Hi Sindibade

MAD is not necesary but recomanded to recover and avoid a split brain.

I'dont use LDAP MAD because I had problems in the past and the third device (connected by LA) has to support LDAP MAD too. I use BFD MAD => it is simple and works great. BFD-MAD uses a third link (i.e. a cheap cupper link) between the IRF members to shut down a possible second master in the IRF.

See the manual for the configuration steps.

br

Manuel

[sindibade_1]

I think that i have the same problem with LACP MAD, because i use the E5500G at the access layer and E5500G LA doesn't support LACP MAD.

but in my case if i use just BFD MAD will using LA on uplinks betwen A7506 and E5500G switchs can't make a problem when a virtual switch split occur?

regards

[manuel.bitzi]

The access switch does not affect the BFD MAD. BFD MAD will detect a split brain and deactivates the second master untill the splitbrain is solved. Then it activates the second irf-member and forces a resync.

br

Manuel

[sindibade_1]

Best regards

[RajRaj]

Implementing BFD MAD and LACP MAD is recommended to avoid split brain scenario.

[mnantel]

The switching fabric is always active on all SRPUs that are not in a failed state, on all members of an IRF pair (and later on an IRF quad) of 7500s.

On a 7506 chassis, each 384Gbps SRPU will provide 24Gbps (half-duplex) of bandwidth per slot. Redundant SRPUs are therefore required to avoid 2:1 oversubscription for 48 port Gbps or quad 10Gbps LPUs.

On a 7510 chassis, each 384Gbps SRPU will provide 12Gbps (HDX) of bandwidth per slot. Each 768Gbps SRPU will provide 24Gbps (HDX) of bandwidth per slot. If you the simple math, you can have anywhere from 4:1 oversubscription with single 384Gbps SRPU up to 1:1 with dual 768Gbps SRPUs.

As a general rule of thumb, don't bother selling SA modules and aim to sell SC and above due to the routing limitations highlighted by another poster. It was suggested to me that LPUs revert to the common lowest denominator when combining module types with regards to table capacities so in any scenario be mindful of your expected requirements in that regard when building 7500 chassis designs.

When answering non-blocking requirements, the 7500 family is to this date going to provide that for 48 ports gig linecards and 4 port 10Gbps linecards. Current backplane architecture (2 XGBUS per slot with each 4 XAUI clocked at 3.125Gbps) does not suggest future capacity upgrade will be possible for this chassis. We have the 9500, 12500 and now the new 10500 for customers with higher 10Gbps density requirements.