Unsolved
This post is more than 5 years old
111 Posts
1
13251
January 6th, 2010 07:00
V-Max Best Practice FA connections
Hi guys,
Just got our new V-Max..
And I wanted to plan the connections for the FAs.
Was wondering what is the best practice for 4 engine (total 64 fiber ports)?
I'm planning to use 4 for SRDF, and 20 for hosts.
In the old Symmetrix we used to follow the "rule of 17", is it still applicable in the V-Max?
Thanks,
Maxim
No Events found!
Quincy561
1.3K Posts
0
March 11th, 2010 08:00
Every drive in a V-Max has the dual initiator function setup across the dual independent DAs in the engine chassis. Every function in the chassis is redundant, power, cooling, memory, etc so this is really no different than a DMX3/4 with one chassis and pairs of DA boards that control all of the drives connected to them.
Quincy561
1.3K Posts
0
March 11th, 2010 08:00
The "engine" is just a chassis, just like the chassis of any array such as DMX. I could say the same thing about a DMX chassis. If the chassis goes "poof" then the drives are not accessible.
beran1
9 Posts
0
March 11th, 2010 08:00
Agreed, but they are IN THE SAME ENGINE. Engine goes "poof", now what?
kensagle
59 Posts
0
March 11th, 2010 08:00
Agreed, an engine going "poof" is like a DMX going "poof", both situations would be very bad, but also very unlikely unless there is a local disaster event in which no hardware can sustain a site disaster (generally speaking).
Sagle
HankDorsett
66 Posts
0
March 12th, 2010 07:00
I know two instances of an engine going "poof" on the same box, both happened during a config change. I asked and re-asked the sysadmin if the storage was still online and he said yes.
What you guys are saying is not what my account team told me when we purchased our VMAX. I've called them to see if they can find me documents on what they told me.
Quincy561
1.3K Posts
0
March 12th, 2010 08:00
Hank, that sort of event has really nothing to do with the hardware redundancy. Also this isn't the right forum to be discussing such events either.
It could be possible to have the same failure in a DMX where a code event could cause all DAs to fail (however unlikely that might be).
pappc
1 Rookie
•
52 Posts
0
March 12th, 2010 08:00
Folks,
Environments can survive an Engine failure on a V-Max.
Two things are required: the first being your host connectivity spanning multiple Engines, ergo the original topic of this thread and the basic answer to the topic; the second being the spanning of your disk groups within the array. If any one of your disk groups solely resides on one Engine, then the logical devices residing in that disk group will also be only residing in that Engine (regardless of RAID protection). If the Engine or both physical directors were to fail, then the array could not access all physical disk access being administered by those two directors, and if you have a disk group that is soley contained in that Engine, then the logical devices will go away as well.
So how to remedy logical volumes going away? Simply span the disk groups they are contained in across multiple Engines.
The other note to make on this is that a failing Engine is extremely rare, as is a dual-initiator failure (both directors touching the same drives) on any of the older Symmetrix models.
Hope this helps,
PSE Chris
mpezeur
40 Posts
0
March 13th, 2010 06:00
Hi, we are in the planning stages with EMC on migrating from a DMX-2 3000 to VMAX-1 with 2x engines in each of our 2x data centers. We were using rule of 17 for dual pathed hosts in the DMX. On the VMAX though we are getting conflicting recommendations from EMC on setting up the dual pathed hosts.
The architect of the bin files, as well as our NJ EMC onsite person are saying that EMC is now using the method of keeping the even #'s FA ports together and odd #'s together, ie a dual hba host would use FA7f0 and 9f0, or 8f0 and 10f0. Two other EMC specialists are our project are saying to use rule of 17 as well. I took the Symm Config Mgmt course for the VMAX functios, and in the class they said that the rule of 17 isn't used either.
We would like to apply the standard that is most commonly used in the field by EMC on the new VMAX. Also we are rolling in an upgrade of switches from brocade 24000 to 48000 as part of the upgrade.
Can someone provide what is recommended, or point to a particular document of best practices on the VMAX?
thank you
HankDorsett
66 Posts
0
March 15th, 2010 09:00
This is what was recomended for my dual engine VMAX.
And on my 3 engine VMAX.
mpezeur
40 Posts
0
March 15th, 2010 10:00
thanks Hank. It looks like our architect came up with a very similar scheme for our 2-engine VMAX as they did for you
what part of the country are you in ?
mpezeur
40 Posts
0
March 15th, 2010 11:00
thanks again Hank
HankDorsett
66 Posts
0
March 15th, 2010 11:00
Nice to see they both agree on it.
I'm outside of Chicago.
JVBowling
1 Rookie
•
32 Posts
0
April 16th, 2010 13:00
Just to clarify,
Do all of the ports associated with FA1 connect to Fabric A and all of the ports associated with FA2 connect to Fabric B?
We are getting a VMAX and I'm trying to decide whether to connect Odd directors to Fabric A and Even directors to Fabric B
OR
All '0' ports to Fabric A and all '1' ports to Fabric B...
Using this method we could still follow the rule of 17...
Thanks,
John
mpezeur
40 Posts
0
April 16th, 2010 15:00
After discussing further, and to consider future growth from 2 -> 3 engines, as well as to avoid hosts being on the same engine, we wound put putting FA 7 and 8 on one fabric (brocade 48k) and FA 9 and 10 on our other fabric (48k). We aren't using all 16 ports, rather 12. Therefore if we add a 3rd engine we will have 4x that can be paired with those on the next engine. Therefore 8g0/1, 9g0/1 would be paired with 5g0/1 and 6g0/1 if we expand to the 3rd engine.
Therefore it may look like we are using rule of 17 be we won't when we expand to the 3rd engine.
hope this helps
JVBowling
1 Rookie
•
32 Posts
0
April 19th, 2010 06:00
If you place Director's 7 & 8 down one fabric and 9 & 10 down the other, then you'll have each Engine connected to a single fabric.
I’m not sure I’m comfortable with that idea and would suspect this would go against best practices?
Maybe I can ask a couple Engineers at EMC World (if I get to go) what their thoughts are as related to Port Assignment and Fabric Connections.
I'm still leaning towards one of the following options:
I'm still on the fence whether to to connect -0 ports to one fabric and -1 ports to the other
or
Odd directors down one and Even directors down the other