# Real multi-node benchmarks. Coordinator on M4M64a (LAN 192.168.2.70); workers on other nodes (several processes per node). coordinator: {alias: M4M64a, lan_ip: 192.168.2.70} port: 50600 designs: # (1) scaling in number of hosts at fixed K = 16 workers: 1, 2, 4, 8, 16 hosts (logistic p = 8, N = 2^22, half-block + 1 Newton) - {label: hosts1, model: logistic, p: 8, N: 4194304, K: 16, splits: [1, 2], q: 1, hessian_mode: exact, repeats: 3, workers_per_node: 16, nodes: [M3U96a]} - {label: hosts2, model: logistic, p: 8, N: 4194304, K: 16, splits: [1, 2], q: 1, hessian_mode: exact, repeats: 3, workers_per_node: 8, nodes: [M3U96a, M3U96b]} - {label: hosts4, model: logistic, p: 8, N: 4194304, K: 16, splits: [1, 2], q: 1, hessian_mode: exact, repeats: 3, workers_per_node: 4, nodes: [M3U96a, M3U96b, M2U64, M4M64b]} - {label: hosts8, model: logistic, p: 8, N: 4194304, K: 16, splits: [1, 2], q: 1, hessian_mode: exact, repeats: 3, workers_per_node: 2, nodes: [M3U96a, M3U96b, M2U64, M4M64b, M4M64c, M5M36, m4mm, m4mn]} - {label: hosts16, model: logistic, p: 8, N: 4194304, K: 16, splits: [1, 2], q: 1, hessian_mode: exact, repeats: 3, workers_per_node: 1, nodes: [M3U96a, M3U96b, M2U64, M4M64b, M4M64c, M5M36, m4mm, m4mn, m6ma, m6mb, m5ma, m4md, m4ml, M2M32c, M4BP48, M4BP36]} # (2) protocol comparison at 16 hosts, equal exponent 7/8: (2,1) vs (1,2) vs (4,0)... plus (3,1), stale/single/quantized Hessians; p = 8 and p = 64 - {label: cmp_r1q2, model: logistic, p: 8, N: 4194304, K: 16, splits: [1], q: 2, hessian_mode: exact, repeats: 3, workers_per_node: 1, nodes: [M3U96a, M3U96b, M2U64, M4M64b, M4M64c, M5M36, m4mm, m4mn, m6ma, m6mb, m5ma, m4md, m4ml, M2M32c, M4BP48, M4BP36]} - {label: cmp_r3q1, model: logistic, p: 8, N: 4194304, K: 16, splits: [1, 2, 4], q: 1, hessian_mode: exact, repeats: 3, workers_per_node: 1, nodes: [M3U96a, M3U96b, M2U64, M4M64b, M4M64c, M5M36, m4mm, m4mn, m6ma, m6mb, m5ma, m4md, m4ml, M2M32c, M4BP48, M4BP36]} - {label: cmp_r4q0, model: logistic, p: 8, N: 4194304, K: 16, splits: [1, 2, 4, 8], q: 0, hessian_mode: exact, repeats: 3, workers_per_node: 1, nodes: [M3U96a, M3U96b, M2U64, M4M64b, M4M64c, M5M36, m4mm, m4mn, m6ma, m6mb, m5ma, m4md, m4ml, M2M32c, M4BP48, M4BP36]} - {label: cmp_r2q1_stale, model: logistic, p: 8, N: 4194304, K: 16, splits: [1, 2], q: 1, hessian_mode: stale, repeats: 3, workers_per_node: 1, nodes: [M3U96a, M3U96b, M2U64, M4M64b, M4M64c, M5M36, m4mm, m4mn, m6ma, m6mb, m5ma, m4md, m4ml, M2M32c, M4BP48, M4BP36]} - {label: cmp_r2q1_single, model: logistic, p: 8, N: 4194304, K: 16, splits: [1, 2], q: 1, hessian_mode: single, repeats: 3, workers_per_node: 1, nodes: [M3U96a, M3U96b, M2U64, M4M64b, M4M64c, M5M36, m4mm, m4mn, m6ma, m6mb, m5ma, m4md, m4ml, M2M32c, M4BP48, M4BP36]} - {label: cmp_p64_r2q1, model: logistic, p: 64, N: 4194304, K: 16, splits: [1, 2], q: 1, hessian_mode: exact, repeats: 3, workers_per_node: 1, nodes: [M3U96a, M3U96b, M2U64, M4M64b, M4M64c, M5M36, m4mm, m4mn, m6ma, m6mb, m5ma, m4md, m4ml, M2M32c, M4BP48, M4BP36]} - {label: cmp_p64_r1q2, model: logistic, p: 64, N: 4194304, K: 16, splits: [1], q: 2, hessian_mode: exact, repeats: 3, workers_per_node: 1, nodes: [M3U96a, M3U96b, M2U64, M4M64b, M4M64c, M5M36, m4mm, m4mn, m6ma, m6mb, m5ma, m4md, m4ml, M2M32c, M4BP48, M4BP36]} - {label: cmp_p64_r2q1_stale, model: logistic, p: 64, N: 4194304, K: 16, splits: [1, 2], q: 1, hessian_mode: stale, repeats: 3, workers_per_node: 1, nodes: [M3U96a, M3U96b, M2U64, M4M64b, M4M64c, M5M36, m4mm, m4mn, m6ma, m6mb, m5ma, m4md, m4ml, M2M32c, M4BP48, M4BP36]} # (3) many partitions per host: K = 64, 128 on 16 hosts (4, 8 processes each) - {label: K64_r2q1, model: logistic, p: 8, N: 4194304, K: 64, splits: [1, 2], q: 1, hessian_mode: exact, repeats: 3, workers_per_node: 4, nodes: [M3U96a, M3U96b, M2U64, M4M64b, M4M64c, M5M36, m4mm, m4mn, m6ma, m6mb, m5ma, m4md, m4ml, M2M32c, M4BP48, M4BP36]} - {label: K64_r1q2, model: logistic, p: 8, N: 4194304, K: 64, splits: [1], q: 2, hessian_mode: exact, repeats: 3, workers_per_node: 4, nodes: [M3U96a, M3U96b, M2U64, M4M64b, M4M64c, M5M36, m4mm, m4mn, m6ma, m6mb, m5ma, m4md, m4ml, M2M32c, M4BP48, M4BP36]} - {label: K128_r2q1, model: logistic, p: 8, N: 4194304, K: 128, splits: [1, 2], q: 1, hessian_mode: exact, repeats: 3, workers_per_node: 8, nodes: [M3U96a, M3U96b, M2U64, M4M64b, M4M64c, M5M36, m4mm, m4mn, m6ma, m6mb, m5ma, m4md, m4ml, M2M32c, M4BP48, M4BP36]} - {label: K128_r1q2, model: logistic, p: 8, N: 4194304, K: 128, splits: [1], q: 2, hessian_mode: exact, repeats: 3, workers_per_node: 8, nodes: [M3U96a, M3U96b, M2U64, M4M64b, M4M64c, M5M36, m4mm, m4mn, m6ma, m6mb, m5ma, m4md, m4ml, M2M32c, M4BP48, M4BP36]}