New to SRDF

I'm neither a cisco expert nor a cisco fan .. but if you already have an IP infrastructure between the two sites and you already are using all the processors on both boxes I think it's easier to buy directly GigE ports for your DMX and connect GigEs to the IP infrastructure. Otherwise if you can migrate the hosts from processor D to other already existing processors, you can ask the Customer Engineer to build a brand new binfile for your boxes and convert the D processor in a RDF processor. Now you have to "translate" the FC protocol in some IP frames .. But it's up to the Cisco Switches such a translation

But I'm pretty sure that your AM will be more then happy to assist you in qualifying your needs and in building a proposal !!

Title: Symmetrix Remote Data Facility (SRDF) Connectivity Guide

Abstract: This document is intended to provide the reader with a connectivity-focused guide for deploying SRDF.

Creation Date: 01-MAY-07
URL: https://powerlink.emc.com/nsepn/webapps/btg548664833igtcuup4826/km/live1/en_US/Offering_Technical/Technical_Documentation/300-003-885_a02_elccnt_0.pdf

I agree with Stefano: migrate the hosts attached to 2 D processors to some other port and have your CE configure these D processors for SRDF (or buy 2 new FC blades for more host connectivity possibilities and use the D ports for SRDF).

I've worked with GigE in DMX and we're now doing FC to Cisco 9216i's and have these Cisco do the FCIP translation and I must say I'm far more happy with this Cisco FCIP implementation. First of all we have weird tcpip issues between both DMX's (the DMX's try to create some sort of port channel / trunk, but since we were using 2 separate links between both DMX's with 2 different subnets, we couldn't get the sum of the available bandwidth working for us (we had 1Gb + 512Mb and the outcome was 2 x 512Mb, so we weren't using 512Mb of the total). When we switches to FC from DMX to Cisco and had the Cisco's do the FCIP (+ data compression !!!!) we were able to use all of the 1,5Gb plus with the compression we can see we sometimes reach as much as 300MB/s (yes, megaBYTES per second and no, these aren't just peek values).

My advice would be: buy new FC cards or migrate hosts away of some FC processors and use the Cisco way of connecting the two sites !

That doesn't surprise me. The load balancing algorithm in SRDF is such that if you have different speed links in the same RDF group, the links will all operate at the slowest speed.

But still: now we're doing SRDF over FCIP we still have the imbalance, but we are doing 1Gb on 1 link and 512 on the other. No, that's not entirily true: we limited the max bandwidth to 800Mb on one line and 400Mb on theother, so theoretically we should be getting 120MBps and I can see that when we're really pushing it, we're getting 350 MBps.

And about the loadbalancing you mentioned: I can see 200 MBps on 1 link and 140 MBps on the other, so it's not evenly balanced at all ! But I must partially agree with you: as long as we don't push really really hard (the links between the 2 DMXs aren't fully loaded) we can indeed see that each RAF is transmitting about the same amount of data per second.

Thanks for all the good info. It now looks like we will be getting two new dmx-4 arrays at each site. I'll try to get two ports on different dmx-4 directors for SRDF.

It sounds like using the MDS to connect sites might be best? If that's true we would order a gigE module for our MDS switches. I'm guessing we would have an orange cable running from the dmx-4 SRDF ports to the gigE ports on the msds switches.

I'm guessing the MDS switches would then connect to our cataylst 6500 switch and then go out through the router from california to texas. In texas we would have a similar configuration where the tx router would connect to the 6500 then to the MDS gigE and finally to the SRDF port in the DMX.

Does this sound correct? I will meet with EMC to discuss all of this but I want to get a basic understanding of the technology so I know what to ask.

Hi Brad,

I think your on the right track...

You will need to install an FCIP capable line card into each of your Cisco MDS chassis and then connect the gig/E interfaces from the MDS line card to your network equipment. These are GigE fibre connectors, so make sure you have spare fibre ports on your 6500 switch.

The IPS-4 and IPS-8 line cards are not recommended as they do software compression and the throughput from those is quite poor in comparison to the combo cards such as the 14+2 or 18+4 which use hardware compression chips and give much higher throughput figures.

The SRDF ports on the Symm are still Fibre channel, so they go into a regular FC port on the MDS.

Once you've got the cards in, its best that you create a new VSAN dedicated for just the SRDF ports, and adjust the trunk allow list on the fcip interface (or port channel if you use multiple ports) so that only the SRDF vsan is permitted to use the link as you do not want to merge your production host fabrics across the IP link.

Do make sure you adjust the trunk allow list before enabling the interface as otherwise you're fabrics will merge and you'll have to spend ages sorting out the combined zonesets.

You then need to create zones for the SRDF directors (RF ports) so that they can see each other via the fabric.

That's basically it from a connectivity perspective however there can be some tuning required to get the best out of the network. A definite recommendation is to set the throughput limit on the interface so that you are not trying to over-drive the link.

Good Luck,

Glen.

Never did it myself and everything I say come from reading only, please correct me if I am wrong,
I think unless you don't have gigE connection between MDS and catalyst now, Then you need a gigE module ( may be IPS-4 or IPS-8 ) for the MDS, It is to connect MDS to the Catalyst switch.

MDS9000 Fabric switches support FCIP, once enable and configure, FCIP is ready and can be use.

Your new DMX (California) use RF port to connect to MDS ( FC protocol ). FC data will be translated to FCIP data inside MDS and pass thru the Ethernet module ( in TCP-IP data stream) to the Catalyst switch. It will in turn send the data stream ( as TCP-IP) to the catalyst switch ( Texas ). The process is reverse at destination... catalyst to MDS ( TCP-IP) and MDS convert the TCP-IP back to FC data frame and feed the FC data to the target DMX in Texas.

The 2 DMX will see they are connected via FC

We have this:
DMX RAF 4D0 -- Cisco 9216i #1 FC1/3 = FC patch cable (50um MM)
DMX RAF 13D0 -- Cisco 9216i #2 FC1/3 = FC patch cable (50um MM)

Cisco 9216i #1 GigE 1 -- network infra = fiber ethernet patch cable (62.5um MM)
Cisco 9216i #2 GigE 1 -- network infra = fiber ethernet patch cable (62.5um MM)

And on the other site the same thing of course. the hardware compression on the Cisco 9216i is great and I expect it to be great as well on the new 9222i.

If you want to use for example port 4D0, you must remember that 4D1 is also an SRDF port now. This is because the whole processor is SRDF enabled and not only 1 port. The same goes for processor 13D in my example. You don't need to do any mapping or masking for symdevs you want to replicate, just do a "createpair" and the DMX does the rest for you.

So for each port you'll have:
DMX "CA" -- FC patch -- Cisco 9216i/9222i -- ethernet patch -- network infra -- routed network -- network infra -- ethernet patch -- Cisco 9216i/9222i -- FC patch -- DMX "TX"

I think we will shoot for the 18/4 MultiService module for each mds 9506. This would give us 18 additional FC ports but more importantly would provide 4 GigE ports for FCIP.

On this 18/4 module we would use two of the 18 fc ports to connect the DMX SRDF ports. We would then use two (or more) of the 4 GigE ports to go to our cataylst for FCIP to the other site in Texas.

That's the way to do it ! The 14+2 module is the old one (2Gbps). These 14+2 or 18+4 modules are the same as the ones for the 9216i/9222i by the way.

Thanks all for the help!

Can you please provide us with feedback in the form of points ? You can "give away" 1 "correct answer" en 2 "helpful answers". This way we know what answer was all right and which not.

I need some help in how to configure RLD for symmetrix DMX-4

we are planning to replicate data from one DMX to other and shift the DMX to the DR site

so can you please tell me the basic hardware configurations and steps I have to follow