SearchStorage.com's newest contributor Arthur Scrimo answered the following user-submitted questions about storage networking. His expert advice addresses several SAN design issues, including how to zone multiple operating systems on a single SAN, how much storage should be put on a pair of HBAs and how to add a switch to one of three production locations.
Question: After reading a recent article about multiple operating systems on a single storage area network (SAN), I was curious about the following:
1. If I use WWN zoning from the HBA on the server to the Fibre Channel interface on the SAN storage array, do I have to be worried about data from different operating systems running through the same interface?
In most cases the answer would be no. However, it is in your best interest to check with storage array vendors to confirm this. The storage, host and HBA vendors produce compatibility matrices that will be very important in storage architectural decisions. It is best to work with these vendors very closely as they often produce updated matrices on at least a monthly basis. HDS's 9980V offers host control domains that apparently let multiple operating systems run along the same FC path.
2. Can the data be corrupted with multiple operating systems and WWN zoning implemented?
No. With the use of LUN masking on the storage array, the SAN zoning method does not matter. You can use your zoning method of choice, but the real key is the LUN masking function that most enterprise storage arrays use. LUN masking allows you to assign a specific LUN or group of LUNs to a specific host HBA. This will allow only the HBA with that WWPN to access that LUN, thus preventing any other host from gaining unauthorized access to your data.
Question: What is the industry standard for the amount of storage that should be put down a pair of HBAs in a SAN? At what point should we look at adding more HBAs to a host due to the amount of dual-attached storage currently configured for that host?
Unfortunately this is a hard question to answer. However, I would say that there are a few factors that can assist in your decision. First, I would say that the amount of storage that can be assigned to a set of HBAs is decided by the operating system and the multi-pathing drivers. As a rule of thumb, I always design HBA connections to support a sustained peak of 70%. This is a safe number that can be used in order to help protect against maxing out an HBA in case of an unusual spike in performance (you can use 80% or ever 90% based on your data profile). What this means is that most of the time the HBA will be in the area of 50% utilization or less, and can peak at about 70% sustained. Once I see the HBAs peaking over 70% sustained utilization, I look at what
factors are driving the increase in performance and go from there. In many cases, the host processes can be coordinated more effectively to move some of the performance load to off-peak hours, but on other hosts you may
have to add HBAs in order to fix your problems.
Question: We have a fabric across three different production locations and we are about to add a switch to one of the locations. Should I configure the switch domain prior to connecting it to the fabric or should I boot up the
switch and accept the switch domain it gets assigned? Is there any conflict if the switch domain is already configured when booting up the switch?
I always manually assign a unique domain ID prior to adding a switch to an existing fabric. Most of the switch vendors have ironed out the issues that were related to fabric switch additions, but I find the manual assignment to be the safest and best approach.
About the author: Arthur Scrimo is a senior SAN architect with IBM ViewPointe.
