difference between a director class SAN switch and a workgroup SAN switch ?

director = more ports, redundancy, bandwidth, protocols, performance, advanced features  and finally more $$$

In terms of Performance, how much I am giving up ?  I understand the LAN switches calculate base on how many packets it can switch, how is SAN switch being measure ? Is there any rule of thumb (or best practice) to choose between director class and workgroup class ?

You have look at each individual vendor (Cisco / Brocade) and look at data sheets for each model.

Your choice may also depend upon the way SAN topology is designed in your Data Center. There is some thing known as a core-edge topology, where you would need a director class switch at core and the departmental switch at the edges. If you have a few devices to connect you may not want to spend on the costly director class switches, you may go for a departmental switch. If you have a big infrastructure to set up then you may need the director class as well. You may also check the details of these switches on Cisco and Brocade websites.

If you need more ports than a single edge can give you, you can choose a director in a collapsed core model, so backend as well as front end connected to your director, which will save on equipment, thus rack space and cables. Furthermore, as Dynamox mentions, a director can house several blades with different features you might need, such as iSCSI, FCIP, encryption, oversubscribed ports, inter VSAN routing (IVR).

So what exactly do you really need ?

RRR,

What I am looking is to replace my current Mcdata ED140M(which is at 2 Gbps/port) , obvious, this is old and (kinda)slow compare to today's technology, currently about 110 fiber connections(including the front end director ports from the symmetrix), fanning ratio is about 7 hosts to 1 Fa port. I am really need is justification to upgrade to a new director (like Cisco MDS9500 series) which is $$$$, and at the same time wondering if I can get away with Cisco MDS9100 series$$, by buying multiple of them to acheive the redundancy at a fraction of the cost ;-)

we are in the process of migrating from ED-140Ms to MDS 9513s.  Yes, you can get a couple of 9148 and create a mash fabric but you will burn so many ports on ISLs that at some point it might be cheaper to get a 9509 (9506 would be too small for you) and just get the number of blades/ports that you need and grow it as you environment grows. For example we bought a couple of blades that are 8G capable, 24 port per blades. These are over-subscribed 2:1 , on these blades we put our storage arrays and big oracle databases that can actually drive some serious data. We also bought a couple of 48 port cards, these are over-subscribed as well but here we put our low performance systems ( dev, qa ). I really enjoy my MDS, rock solid platform.

How many storage ports do you need to connect ?

Do you need anything else than FC connectivity, like iscsi, fcip ?

Do you need multiple VSANs and routing between them ?

Would oversubscribed 4Gb ports be ok or do you need true line rate performance on 4 or even 8Gb ?

Do you need some free ports to be able to grow ?

I don't know anything about pricing, but to get 110 ports, we might consider two 9148s and one 9124 which gives you 2 x 48 + 24 = 120 ports, but you'll need some ports on each switch for ISLs. The 9124 is line rate, the 9148 I'm not sure about, but I think it's line rate at 2Gb and 1:2 on 4Gb, but I'd need to check that somewhere as I've never seen a 9148 yet. And then you'd need two of these setups, one for each fabric.

Suppose the 1 x 9124 as "core" and the two 9148s are ok for performance, you could have 112 ports, because from the 9124 you could implement a portchannel of two ISLs, taknig 2 ports from each 9148 and 2 x 2 from the 9124, leaving you with 120 - 4 x 2 = 112 ports. If you need some free ports for future growth, you might want to consider 3 x 9148.

How many storage ports do you need to connect ?

Fa ports ? 12 to 14.

Do you need anything else than FC connectivity, like iscsi, fcip ?

no iscsi, no fcip (but will come into play as we might move to Recoverpoint applicance for replication.). For now FC with npiv capbability.

Do you need multiple VSANs and routing between them ? no

Would oversubscribed 4Gb ports be ok or do you need true line rate performance on 4 or even 8Gb ?

what do you mean, currently I have about 7 to 10 servers(2Gbps) , sharing(zoning to) one Fa port (2Gbps) ! been running OK, like to keep it the same but 7 to 10 (4Gb) to single Fa port (4Gb)

Do you need some free ports to be able to grow ? Yes.

If you need free ports for future growth you might want to consider getting 3 9148s per fabric right away, but even that will only give you about 24 free ports, since we already saw that in my assumption you'd need two 9148s and one 9124 to get what you would need at this moment.

I don't know what these wonderful things cost, but when you already need three 9148s and maybe even 1 more, buying a single 9509 with room for 7 blades might be cheaper. And as Mr Dynamox already mentioned, ISLs take up ports on both switches, so 1 portchannel consisting of 2 ISLs will cost 2 x 2 ports on the switches the portchannel is defined on. Having a single 9509 (per fabric) will save on the ISLs !

If you have a 9509, you'll get room for 7 blades. I'd suggest 2 non oversubscribed blades for FA connectivity (12 ports each) which will leave you with 2 x 12 - 12 (or 14) = 12 (or 10) free ports for FAs or ISLs to other switches. The remaining 5 blades can be used for 12 port, 24 port or 48 port blades giving you 5 x 12 = 60, 5 x 24 = 120 or 5 x 48 = 240 host ports. Or a mix if you like. The 24 port blades are 2:1 oversubscribed and the 48 ports blades 4:1 in terms of bandwidth. I'd go for the 24 or 48 port blades, since you come from a 2Gb environment, which translates to a 2:1 oversubscribed 24 port blade. If you measure the performance on your current SAN ports, do all ports really need to be 2Gb at this moment or would an average of 1Gb be sufficient ? Brocade SAN Health can be used to get a nice graph of the last 48 hours at the most. You can easily see if you reach the maximum. Besides: 2 separate 2Gb ports guarantee that each port can actually handle 2Gb, but having a 2 4Gb ports which are 2:1 oversubscribed can each get 4Gb, but will share the 4Gb bandwidth when needed.

Does this make sence ?

Directors aren't really that much more expensive than the departmental ones.  The list price is a lot more, but usually you can negotiate a price pretty close to the price of the workgroup switches.

If you knew what I bought our last pair of DCX's for, you'd probably yell at your sales rep.

If you want a director and need new switches anyway, I suggest you talk to your sales rep and just see what they can do.  They might be able to surprise you.  Brocade and Cisco both like to sell directors more than they like to sell departmental switches.  They have more "staying power" than small switches.

If you anticipate a lot of growth any time soon - I suggest you go with a director.  100 hosts isn't a whole lot - but it's starting to get to the point where a director would make sense.  If you anticipate expanding in the future, you'll start to see more and more growth problems as you keep investing in departmental switches.

driskollt wrote: Directors aren't really that much more expensive than the departmental ones.  The list price is a lot more, but usually you can negotiate a price pretty close to the price of the workgroup switches. If you knew what I bought our last pair of DCX's for, you'd probably yell at your sales rep.

i do not know about DCX but Cisco Director class are much much more expensive than departmental switches. I can get a MDS 9134 for $10k ..you will probably pay 10k just for the sticker on your director that says Cisco on it.

Thank EU

$10k for a 9134 ? I'll take 2 ! I've seen prices a lot higher than that... I think....

Gents,

Thanks for all the analysis you've done. I think eventually I will need less ports because of CISCO's UCS, and "HP's virtual connect" that my place will move to for the windows servers. In that case, probably 4+ hosts are using sharing 1 8GB fiber port. Taking the number of ports needed out of the consideration, director will seem kinda overkill, but will the MDS9148s able to handle the traffic. tough decison !