Question: Scaling Feedsim Dataset Size

[vpetrucci]

What would be a reasonable strategy for scaling the Feedsim dataset to leverage large-memory systems, such as one with 256 cores and 2 TB of RAM (i.e., 8 GB per core)?

Specifically, would it be appropriate to increase parameters like "graph_scale" and/or "num_objects" to better utilize the available memory?

Also, what are suggested values for parameters that scale the workload meaningfully without distorting its intended behavior or characteristics?

Starting leaf node service

monitor_port=$((port-1000))
MALLOC_CONF=narenas:20,dirty_decay_ms:5000 build/workloads/ranking/LeafNodeRank \
    --port="$port" \
    --monitor_port="$monitor_port" \
    --graph_scale=21 \
    --graph_subset=2000000 \
    --threads="$thrift_threads" \
    --cpu_threads="$ranking_cpu_threads" \
    --timekeeper_threads=2 \
    --io_threads="$EVENTBASE_THREADS_DEFAULT" \
    --srv_threads="$SRV_THREADS_DEFAULT" \
    --srv_io_threads="$srv_io_threads" \
    --num_objects=2000 \
    --graph_max_iters=1 \
    --noaffinity \
    --min_icache_iterations="$icache_iterations" &