Load testing of iSCSI shared storage for Hyper-V 2012 R2 environment

I have a brand new datacenter environment that consists of 4 physical servers and an iSCSI storage array (plus switches, etc). The 4 servers are destined to become Hyper-V 2012 R2 hosts and will host a variety of virtualised server workloads (no VDI) that are due to be migrated from a VMware cluster.

I need to benchmark the storage before putting it in production. I'll probably use IOmeter but will consider other recommendations. I've found a couple of good resources such as this MSDN blog post detailing IO profiles for IOmeter, but I need some guidance around the specifics of how to actually put together a worthwhile test plan. My questions are:

1. How long should each test run for?
2. Should I test all 4 hosts concurrently (and somehow aggregate or average the results), or test each host one by one?
3. Should I run IOmeter on the physical host, or from from within a VM (against a virtual disk)?
4. How do I create a write test that will saturate all levels of caching, thus revealing the actual disk performance (the array controller has a cache and I believe Hyper-V automatically caches writes)?

Thanks in advance!