What happens to a VM that is hosted on a VMFS datastore when you physically remove both of the mirrored drives from the array? Kernel panic/blue screen right?
What if you were replicating that LUN to another site using array based replication? Would the VM still crash?
Not if you’re using Symmetrix Remote Data Facility, Synchronous mode (SRDF-S).
Let’s rewind a bit. One of the main architectural attributes that distinguish an enterprise array from a mid-range array is the connectivity and cache model. In a midrange array you will typically see an active/passive controller from a LUN path ownership perspective, while an enterprise array provides an active/active controller model for because the front-end adapters don’t “own” the LUN per-se (Yes, I’m oversimplifying this a wee bit, but only to skip past this detail and get to my point!).
As you can see in the simplified images, enterprise arrays provide a complete cache abstraction layer between front end controllers (host side) and back end controllers (drive side). In other words, if you want to send a command or piece of data from the host to the drive, you must do so through cache.
This kind of abstraction, much like the server abstraction provided by VMware ESX, is a key enabler for much of the genius that occurs in all shared storage platforms.
In the world of Symmetrix, a mirror-position is a device in the system that a segment of cache refers to. When a write occurs, it is acknowledged by cache, then the system will deliver the write to all of the necessary mirror-positions. If it is a “2 Way Mirror” (RAID1) it will write to mirror-position 1 (first RAID1 device) and mirror-position 2 (second RAID1 device).
When you create a RAID1 configuration on a Symmetrix DMX or VMAX the system refers to it as a “2-Way-Mir” (two way mirror). That is the systems way of saying that each cache segment responsible for staging IO within this volume has two positions to write to – each of which are on a physical device (this architecture allows RAID1 volumes to stretch far beyond the capacity limits of a single drive – neato!) (By the way, with Virtual Provisioning technology implemented in the VMAX, the virtual pool’s device is the mirror position, and the pool takes care of RAID – the idea here is still the same though).
What if you wanted a “3-Way-Mir”? Well technically, that’s what SRDF-S is; it’s a three-way mirrored volume. The first two mirror positions are local devices, and the third is a device on the remote array.
So what happens when you simultaneously remove both of the drives in a two drive two way mirror with SRDF-S configured?
Your VMs stay online. Read latency will suffer as reads are now being serviced by the remote site, but writes are the same speed because they were limited by the inter-site latency anyway.
How awesome is that?