seu-3dgs / code /survival.py
Lightcap's picture
Upload folder using huggingface_hub
121e1fb verified
Raw
History Blame Contribute Delete
6.98 kB
"""E13: survival / reliability model (runs on CPU, e.g. the local M4).
From the per-bit catastrophe probability and the accumulated-dose sweep we model
the probability that a rendered frame is catastrophic after k independent upsets
as 1-(1-p_c)^k, validate it against the multi-upset measurements, and translate
it into a mean time between catastrophic frames for a model of B stored bits under
representative single-event-upset rates (ground, avionics, low-Earth orbit), with
and without the support guard.
"""
import argparse
import glob
import json
import math
import os
import numpy as np
import matplotlib
matplotlib.use("Agg")
import matplotlib.pyplot as plt
plt.rcParams.update({"font.family": "serif", "mathtext.fontset": "cm", "font.size": 12,
"axes.labelsize": 12, "legend.fontsize": 10, "lines.linewidth": 1.9,
"lines.markersize": 5.5, "axes.grid": True, "grid.alpha": 0.25,
"savefig.dpi": 220, "savefig.bbox": "tight"})
# representative SEU rates (upsets per stored bit per hour), order-of-magnitude,
# from terrestrial/avionic/space soft-error literature.
SEU_RATE = {"ground": 1e-12, "avionics": 3e-10, "LEO": 1e-8}
CAT_PSNR = 25.0 # a frame is catastrophic if global PSNR falls below this
def per_bit_catastrophe(root):
"""p_c from the main campaign: fraction of uniform-random single-bit upsets
that are catastrophic (footprint > 1% or non-finite), weighted over fields."""
import numpy as np
shards = glob.glob(os.path.join(root, "campaign", "shard_*_fp32.npz"))
shards = [s for s in shards if not s.endswith("_guard.npz")]
fr, cat = [], []
for s in shards:
d = np.load(s, allow_pickle=True); a = d["data"]; cols = list(d["cols"])
ci = {c: i for i, c in enumerate(cols)}
fr.append(a[:, ci["fracchg"]]); cat.append(a[:, ci["cat"]])
if not fr:
return None, None
fr = np.concatenate(fr); cat = np.concatenate(cat)
is_cat = (cat > 0.5) | (fr > 0.01)
p_c = float(is_cat.mean())
# guarded residual, if present
g = glob.glob(os.path.join(root, "campaign", "shard_*_fp32_guard.npz"))
pg = None
if g:
frg, catg = [], []
for s in g:
d = np.load(s, allow_pickle=True); a = d["data"]; cols = list(d["cols"]); ci = {c: i for i, c in enumerate(cols)}
frg.append(a[:, ci["fracchg"]]); catg.append(a[:, ci["cat"]])
frg = np.concatenate(frg); catg = np.concatenate(catg)
pg = float(((catg > 0.5) | (frg > 0.01)).mean())
return p_c, pg
def multiupset_pcat(root):
"""empirical P(catastrophic frame) vs k from the no-guard multi-upset sweep."""
per_k = {}
for fp in glob.glob(os.path.join(root, "multiupset", "multiupset_*_fp32.npz")):
if fp.endswith("_guard.npz"):
continue
d = np.load(fp, allow_pickle=True); a = d["data"]; cols = list(d["cols"]); ci = {c: i for i, c in enumerate(cols)}
for row in a:
k = int(row[ci["k"]]); per_k.setdefault(k, []).append(row[ci["psnr"]])
ks = sorted(per_k)
pcat = {k: float(np.mean(np.array(per_k[k]) < CAT_PSNR)) for k in ks}
return ks, pcat
def fmt_hours(h):
if h <= 0:
return "n/a"
yr = h / 8760.0
if yr >= 1e5:
e = int(math.floor(math.log10(yr)))
m = yr / 10 ** e
return f"$\\sim{m:.0f}\\times10^{{{e}}}$ yr"
if yr >= 10:
return f"{yr:,.0f} yr".replace(",", "{,}")
if yr >= 1:
return f"{yr:.1f} yr"
d = h / 24.0
if d >= 1:
return f"{d:.0f} d"
return f"{h:.1f} h"
def main():
ap = argparse.ArgumentParser()
ap.add_argument("--root", default="data_local")
ap.add_argument("--out", default="../generated")
ap.add_argument("--bits", type=float, default=2.55e8, help="stored bits in the model")
args = ap.parse_args()
os.makedirs(args.out, exist_ok=True)
p_c, p_g = per_bit_catastrophe(args.root)
ks, pcat = multiupset_pcat(args.root)
macros = {}
if p_c is None:
p_c = 1e-3
macros["pcUpset"] = f"{p_c*100:.3f}"
macros["pcGuard"] = (f"{p_g*100:.4f}" if p_g is not None else "0.0000")
B = args.bits
macros["modelBits"] = f"{B/1e6:.0f}\\times10^6"
# survival figure: empirical P(cat|k) vs the 1-(1-p_c)^k model
plt.figure(figsize=(6.2, 4))
if ks:
ke = np.array(ks)
plt.plot(ke, [pcat[k] for k in ks], "o", label="measured")
kk = np.logspace(0, np.log10(max(ks)), 100)
plt.plot(kk, 1 - (1 - p_c) ** kk, "-", label=r"$1-(1-p_c)^k$ model")
plt.xscale("log"); plt.xlabel("simultaneous single-bit upsets $k$")
plt.ylabel("P(catastrophic frame)"); plt.legend(); plt.grid(alpha=0.3)
plt.savefig(os.path.join(args.out, "fig_survival.pdf"), bbox_inches="tight"); plt.close()
# reliability table -> macros (MTBF for first catastrophic frame).
# LaTeX command names must be letters only, so use camelCase keys.
NM = {"ground": "Ground", "avionics": "Avionics", "LEO": "Leo"}
for env, rate in SEU_RATE.items():
ev_per_hr = rate * B * p_c
mtbf = float("inf") if ev_per_hr <= 0 else 1.0 / ev_per_hr
macros[f"mtbf{NM[env]}Ng"] = fmt_hours(mtbf)
ev_g = rate * B * (p_g if p_g else 1e-9)
mtbf_g = float("inf") if ev_g <= 0 else 1.0 / ev_g
macros[f"mtbf{NM[env]}G"] = fmt_hours(mtbf_g)
# emit a small table and macro file
with open(os.path.join(args.out, "tab_survival.tex"), "w") as f:
f.write("\\begin{table}[tbp]\n\\centering\n")
f.write("\\caption{Estimated mean time between catastrophic frames for a "
"model of $\\modelBits$ stored bits under representative single-event-upset "
"rates, without and with the support guard. Rates are order-of-magnitude "
"values from the soft-error literature.}\n\\label{tab:survival}\n")
f.write("\\begin{tabular}{lrr}\n\\toprule\nEnvironment & no guard & support guard \\\\\n\\midrule\n")
names = {"ground": "ground (sea level)", "avionics": "avionics ($\\sim$10 km)", "LEO": "low-Earth orbit"}
NM = {"ground": "Ground", "avionics": "Avionics", "LEO": "Leo"}
for env in ["ground", "avionics", "LEO"]:
f.write(f"{names[env]} & {macros['mtbf'+NM[env]+'Ng']} & {macros['mtbf'+NM[env]+'G']} \\\\\n")
f.write("\\bottomrule\n\\end{tabular}\n\\end{table}\n")
with open(os.path.join(args.out, "survival_numbers.tex"), "w") as f:
# provide safe defaults too
defaults = {"pcUpset": "0.000", "pcGuard": "0.0000", "modelBits": "2.6\\times10^8"}
for nm in ["Ground", "Avionics", "Leo"]:
defaults[f"mtbf{nm}Ng"] = "n/a"; defaults[f"mtbf{nm}G"] = "n/a"
for k, v in defaults.items():
macros.setdefault(k, v)
for k, v in macros.items():
f.write(f"\\newcommand{{\\{k}}}{{{v}}}\n")
print("SURVIVAL macros:", macros)
if __name__ == "__main__":
main()