Skack All American 31140 Posts user info edit post |
http://www.amazon.com/gp/product/0072130725/qid=1135024025/sr=1-8/ref=sr_1_8/104-7320529-1307905?s=books&v=glance&n=283155
This was a pretty good book for the nitty gritty of how they work, but I thought it lacked the practical application side a little bit. i.e., They spend too much time telling how iSCSI packets are encapsulated and not enough time telling you how to build a SAN.
^ Yep...Redbooks.ibm.com has a lot of resources. Just try searching for "SAN" and "DS4000" (DS4000 is their midrange line of fibre storage.)
Have anything more specific that you want to ask about?
[Edited on December 19, 2005 at 3:34 PM. Reason : s] 12/19/2005 3:29:38 PM |
Skack All American 31140 Posts user info edit post |
Skack's Fibre Channel SAN Primer (because I'm drunk and have little better to do):
Servers: Install a Host Bus Adapter (HBA) in the PCI slot (or two for redundancy). You can run more than that, but it gets complicated, expensive, and sucky. The HBA acts like an ethernet controller in that it just passes data onto the network with no real logic built in (i.e. no raid striping at this level.) In this case it just happens to be passing disk I/O onto the network instead of ethernet traffic. You'll need a multipath driver in the OS if you run more than one HBA. The data from the HBA is pushed via a fiber optic cable directly to the storage or to a switch.
....If directly to the storage, it will work in an "arbitrated loop" (FC-AL) configuration. Most storage controllers will support 1, 2, or 4 connections which gives you access to one server with one HBA, one server with two HBAs (redundant paths,) two servers with one HBA, or two servers with two HBAs (redundant.) The primary limitation here is the number of servers that can connect to the storage.
...If to a fibre channel switch, it will work in a "switched fabric" configuration. The primary exception being when it is connected to a managed hub which creates an FC-AL, but provides connectivity to more hosts than would normally be supported in a FC-AL config. It is not uncommon to have two switches in the SAN therefore creating two completely separate paths to the storage.
Storage Controller - The device that provides all the logic for RAID striping, lun masking (mapping RAID arrays to a single server so that only that server can see it), etc. The storage controller may have built in HDD slots allowing it to accept HDDs, or it may require JBODS (acronym for "just a bunch of disks" i.e. Boxes filled with HDDs, but no logic) to be connected in order to function. Some storage controllers have some level of internal storage, but also accept JBODs to provide additional storage.
- Most entry level and midrange storage controllers will come with a single controller and an optional second controller for redundancy. The controllers arbitrate ownership of the disks. Technically, only one controller can own a RAID array at a time, but load balancing can be done by splitting ownership of multiple arrays between the two controllers so that they have a somewhat even load. In the case of a controller failure or a path failure the functioning controller should grab ownership of all drives and provide access through the functioning path to the server.
- Fibre Channel switches generally have a feature similar to ethernet VLANs which is called "VSANS" and creates separate fabrics. More popular that VSANs (from what I've seen at least) is "zoning." Zoning can be done based on the worldwide name (WWN, similar to a MAC address, but it is built onto every device in the SAN from the HBAs to the storage controllers to each port on the switch) level or at the port level by saying "this WWN can see this WWN" or "this port can see this port." These are called hard and soft zoning.
- Lun Masking (previously mentioned) is done at the controller level and allows only one server (or cluster of servers) to see a RAID array by transmitting information about that array only to the worldwide name (WWN, think MAC address in ethernet terms) of the HBAs in the approved server. Basically, if LUN masking weren't set up and another server saw that array it would try to write a disk signature and screw it all to hell.
These are all just basic, best practices type info. There is an exception to almost everything I've said here, but this is what you'll see ninety some percent of the time. It can get a heck of a lot more complicated than this, but I'm trying to keep it simple. 12/20/2005 11:57:27 PM |