Add memory info to the systems

• Test to make sure they get stored in caliper metadata after experiment runs - look here.
• What can you automatically extract from hwloc info? Start with @amroakmal's script and write a helper script
• There may be more information in /etc/node_info.json on some of the LC machines

Units and how to organize the data

• Use the same units as hwloc, add them in the name of the variable, e.g., sys_cpu_L1_KB_per_core, sys_mem_GB_per_node
• check that the units make sense for the cheapest/smallest CPU on AWS (we can't do fractions)
• If hwloc is reporting X L1s per node, we should also - so update our variables to sys_cpu_L1_KB_per_core (we already have sys_num_cpu_cores) and `

Types of counters

• sys_mem_GB_per_node
• sys_cpu_mem_GB_per_node (same as sys_mem_per_node in CPU-only)
• sys_gpu_mem_GB_per_node (if separate from CPU memory)
• sys_cpu_L1_KB_per_core
• sys_gpu_L1_KB_per_device
• sys_gpu_num_L1_per_node
• sys_cpu_L2_MB_per_node
• sys_gpu_L2_MB_per_node
• sys_gpu_num_L2
• sys_cpu_L3_MB_per_node
• sys_gpu_L3_MB_per_node
• sys_gpu_num_L3
• sys_num_nodes_per_rack
• sys_num_racks
• sys_network_topology (what is available in hwloc on this?)

If you are finding that the counter names are not a good match for what we have on CPU clusters and on El Cap, lets discuss here