cancel
Showing results for 
Search instead for 
Did you mean: 
cancel
1810
Views
0
Helpful
17
Replies

Single or dual 4500e for redundancy

arrkerr
Level 1
Level 1

I apologize if this has been asked before, but I haven't been able to find a solid answer when searching.

We're deploying a new datacenter and will have 20 racks of servers, mostly 1u.  Given our bandwidth needs which are reasonably modest we're planning on using dual 2960s switches at the top of the racks and aggregating on a single or dual 4500e using 10gig links.  The 4500e(s) will also have a bunch of 1gb devices such as firewalls, routers, and load balancers connected to them.  Pretty typical network core.    Given the need for 40 10gb links to racks and then another 20-or-so 1gb, the 4500e chassis seems to fit the bill.

My question is, do we go with a single 4507e with full redundancy (dual 10gb cards, one link to each from each rack, dual 1gb card, and dual supervisors) or a pair of either 4506e switches each with their own cards.  With a virtual chassis the 4506e is appealing, but I'm not up to speed on any limitations that might impose.   With IISU and the like does having two physical chassis vs everything in a single actually buy us anything if they're going to be in the same physical proximity anyway?  A big reason to go with the single chassis is space - 11 rack units vs 20.

If we do go with a single chasis, other than Cisco's HA docs on the 4500e, is there any documentation or case studies that I could use to sell the idea to management?  A number of people, especiailly those in favor of a ton of cheap netgear switches, argue that a single chassis is a single point of failure and we should never do it.

Thanks!

17 Replies 17

canero
Level 1
Level 1

Hello,
Providing chassis redundancy may be important as if you have problem woth the single 4500 chassis ( fan, power supply etc.) you lose all 20 racks pf servers.
you also get more io module slots for future enhancements as well
Best regards

Sent from Cisco Technical Support iPhone App

Fair enough, but doesn't the single chassis have redundant fan and power supply modules?  I guess thats what I'm getting at.

Leo Laohoo
Hall of Fame
Hall of Fame

Wow, 20 racks of server ... I do not believe 2960S can support the massive throughput from a rack-full of servers.  You will/may see some input/output drops.

If you use Sup7E and a 4500R+E, each slot has a backplane of 48 Gbps.   The 4500R+E chassis is ideal if you want to terminate alot of 1 Gbps and a few 10 Gbps.

2960S is a "desktop" access switch.  The 3560X, 3750X or the 3850 can possibly support your servers.

I don't believe this is correct.  The 2960s and 3750x both have the same switching fabric bandwidth and the same forwarding rates.  The 3750x supports more cool features like better stacking and ISSU, but to get that you need IP services which adds a huge cost at TOR for a single feature - a real bummer.

If you have data to show otherwise, please share.  Maybe you're thinking of the older 2950 (non-s) models.

arrkerr
Level 1
Level 1

I'm actually thinking now that we might be better off with two 4500-x switches at the core, and instead of connecting our 1gb ASAs/ASRs/F5s directly to them hang dual 2960s switches off of the 4500-x (10gb uplink) to aggregate.  We'd then be treating our "network services" rack just as we would a normal data rack with a top of rack agg switch.

I'd add that ideally we'd ideally be using a Nexus 6001 with 2248T extenders, but that is most likely way out of the ballpark in both price and capacity.  Awaiting pricing though.

Joseph W. Doherty
Hall of Fame
Hall of Fame

Disclaimer


The  Author of this posting offers the information contained within this  posting without consideration and with the reader's understanding that  there's no implied or expressed suitability or fitness for any purpose.  Information provided is for informational purposes only and should not  be construed as rendering professional advice of any kind. Usage of this  posting's information is solely at reader's own risk.

Liability Disclaimer

In  no event shall Author be liable for any damages whatsoever (including,  without limitation, damages for loss of use, data or profit) arising out  of the use or inability to use the posting's information even if Author  has been advised of the possibility of such damage.

Posting

As you've described, the problem with any single chassis, the chassis itself can become a single point of failure even when loaded with redundant components.  Independent chassis (e.g. VSS, StackWise) provide chassis redundancy although it does raise a question on software redundancy, i.e. multiple chassis are running some form of tightly coupled chassis OS vs. standalone chassis.  Standalone chassis should be, in theory, the most redundant but they too create issues such as load sharing and asymmetrical path issues (e.g. unicast flooding).  If you understand and trust a multi-chassis OS, I personally think it's often the "better" redundancy option.

I don't know if VSS on the 4500s support it, but 6500 VSS supports quad supervisiors, so if you lose a single sup on those you don't lose one of your VSS chassis pair.

Regarding your later post on using a 4500-X VSS pair, yes that might be an idea core for high density 10g.  For twenty 10g ports each, you would need either the 16 port with the optional 8 port module or the 32 port model.

Depending on your remote rack setups, you might also consider 3750-Xs, stacked, in lieu of multiple (individual cabinet) ToR 2960S pairs.  StackWisePlus is a much better stacking technology then FlexStack.  Yes, the 3750-Xs are more expensive, but you might need less if you can bring multiple racks to the same 3750-X stack.  (Depending on how many downstream stacks you actually need, you might also reduce your need for 10g ports on the core.)  Depending on your feature needs, you might even be able to use the LAN Base models which in the later IOSs also support StackPower.

10g is nice, but it's also expensive.  When working with switches within 100m, don't overlook the possibilities of gig copper Etherchannels.  For example compare total cost of 8xgig (copper) build-in ports vs. single 10g (fiber) especially if special module is required.

What am I talking about????

Why am I still harking about the 3560X and 3750X?

I should start recommending people consider the 3850. 

Disclaimer


The Author of this posting offers the information contained within this posting without consideration and with the reader's understanding that there's no implied or expressed suitability or fitness for any purpose. Information provided is for informational purposes only and should not be construed as rendering professional advice of any kind. Usage of this posting's information is solely at reader's own risk.

Liability Disclaimer

In no event shall Author be liable for any damages whatsoever (including, without limitation, damages for loss of use, data or profit) arising out of the use or inability to use the posting's information even if Author has been advised of the possibility of such damage.

Posting

I agree where 3750-X might be considered, so should a 3850.

The problem with the 3750-x is that they require IP Base to do ISSU, which is a huge additional cost.  I'd love to stack (or virtual stack) my top of rack switches, but I have no need for any of the other features and quite frankly I can't justify the addional cost fo rthe single feature.  I love StackPower as well, it is really awesome.  But again, I can't justify the cost.

arrkerr wrote:

The problem with the 3750-x is that they require IP Base to do ISSU, which is a huge additional cost.  I'd love to stack (or virtual stack) my top of rack switches, but I have no need for any of the other features and quite frankly I can't justify the addional cost fo rthe single feature.  I love StackPower as well, it is really awesome.  But again, I can't justify the cost.

ISSU isn't available on the 3750-X or 3850 series. It's for bigger switches and routers. The closest 3750-X feature is the 'Rolling Stack Upgrade' released in 12.2(58)SE, but you need to be careful how you uplink in order to maintain connectivity to the stack. It's available in 'LAN Base' IOS: -

http://www.cisco.com/en/US/docs/switches/lan/catalyst3750x_3560x/software/release/12.2_58_se/configuration/guide/swstack.html#wp1295104

Features quite often move down between feature sets over time. You'd need to check, but I am pretty sure StackPower had dropped down to 'LAN Base' IOS.

I agree with all previous comments about the 2960-S series being used as ToR switch. You might get away with the 3850 series. Certainly, the 3750-X has larger default egress buffers than 2960-S.

Dare I say it, but if you're struggling to spec an appropriate Cisco solution within your budget, then it might be worth considering alternative manufacturers. Better that than provisioning the wrong Cisco equipment and experiencing issues. There is also the 2nd user market or Cisco's own refurbished kit.

shillings: thanks for the correction about RSU and not ISSU.  According to this page RSU is only available in IP Base though, which adds a significant cost to a TOR switch.  Is this page wrong, can you really do it with LAN Base?

http://www.cisco.com/en/US/prod/collateral/switches/ps5718/ps10745/product_bulletin_c25-658743_ps7078_Products_Bulletin.html

arrkerr wrote:

shillings: thanks for the correction about RSU and not ISSU.  According to this page RSU is only available in IP Base though, which adds a significant cost to a TOR switch.  Is this page wrong, can you really do it with LAN Base?

http://www.cisco.com/en/US/prod/collateral/switches/ps5718/ps10745/product_bulletin_c25-658743_ps7078_Products_Bulletin.html


Just checked Feature Navigator http://tools.cisco.com/ITDIT/CFN/jsp/index.jsp

It states that 3750-X 'Rolling Stack Upgrade' is available in 'UNIVERSAL (LAN BASE)' image with either 12.2(58), 15.0(1) or 15.0(2). The new 3850 series is not yet listed.

If you're able to access Feature Navigator yourself, then just note that it won't find 'Rolling Stack Upgrade' when you search for the keyword 'Rolling', but will find it if you search for 'Stack'

Disclaimer

The  Author of this posting offers the information contained within this  posting without consideration and with the reader's understanding that  there's no implied or expressed suitability or fitness for any purpose.  Information provided is for informational purposes only and should not  be construed as rendering professional advice of any kind. Usage of this  posting's information is solely at reader's own risk.

Liability Disclaimer

In  no event shall Author be liable for any damages whatsoever (including,  without limitation, damages for loss of use, data or profit) arising out  of the use or inability to use the posting's information even if Author  has been advised of the possibility of such damage.

Posting

Yes, indeed at I've noted, they are more expensive.  However, again, if you can aggregate multiple racks into a single stack you often need less stack members.  I..e the freed ports per ToR, aggregated, become no longer needed switches.

Regarding ISSU, if we're really using these units a logical stack extenders, how often will you really need to upgrade them?  Sure, it's a nice feature, but we've all used devices without ISSU for years, right?  Of course, if this feature is a critical requirement, then it is, but you might want to weigh it against other features.  We're usually always stuck with some kind of trade-offs.

The newer 3850 offers a huge increase in stack ring bandwidth.  So much so, I would be less concerned about deeper stacks.  It also makes me wonder about their possible usage in a small core; assuming you used MEC gig copper Etherchannel.