power / gui /pages /cpu.py
Deepakresearcher's picture
New build
b555842
Raw History Blame Contribute Delete
5.25 kB
"""
gui/pages/cpu.py — ▣ CPU & C-States
Render function: render(ctx)
ctx keys: ov, runs, tax, lin, age, avg_lin_j, avg_age_j, tax_mult,
plan_ms, exec_ms, synth_ms, plan_pct, exec_pct, synth_pct
"""
import subprocess
import streamlit as st
import pandas as pd
import plotly.express as px
import plotly.graph_objects as go
from plotly.subplots import make_subplots
from gui.config import PROJECT_ROOT, DB_PATH, LIVE_API, WF_COLORS, PL
from gui.db import q, q_safe, q1
from gui.helpers import fl, _human_energy, _human_water, _human_carbon, _gauge_html, _bar_gauge_html
try:
import requests as _req
_REQUESTS_OK = True
except ImportError:
_REQUESTS_OK = False
class _req:
@staticmethod
def get(*a, **kw): raise RuntimeError("requests not installed")
try:
import yaml as _yaml
_YAML_OK = True
except ImportError:
_YAML_OK = False
def render(ctx: dict):
ov = ctx["ov"]
runs = ctx["runs"]
tax = ctx["tax"]
avg_lin_j = ctx["avg_lin_j"]
avg_age_j = ctx["avg_age_j"]
tax_mult = ctx["tax_mult"]
plan_ms = ctx["plan_ms"]
exec_ms = ctx["exec_ms"]
synth_ms = ctx["synth_ms"]
plan_pct = ctx["plan_pct"]
exec_pct = ctx["exec_pct"]
synth_pct = ctx["synth_pct"]
lin = ctx["lin"]
age = ctx["age"]
st.title("CPU & C-State Analysis")
cstate_df = q("""
SELECT e.provider, r.workflow_type,
AVG(cs.c1_residency) AS c1, AVG(cs.c2_residency) AS c2,
AVG(cs.c3_residency) AS c3, AVG(cs.c6_residency) AS c6,
AVG(cs.c7_residency) AS c7,
AVG(cs.cpu_util_percent) AS util,
AVG(cs.package_power) AS pkg_w,
COUNT(cs.sample_id) AS samples
FROM cpu_samples cs
JOIN runs r ON cs.run_id = r.run_id
JOIN experiments e ON r.exp_id = e.exp_id
GROUP BY e.provider, r.workflow_type
""")
if not cstate_df.empty:
st.markdown("**C-State Residency** — higher C6/C7 = deeper sleep = more efficient idle")
CSTATE_COLORS = {"C0":"#ef4444","C1":"#38bdf8","C2":"#3b82f6",
"C3":"#a78bfa","C6":"#22c55e","C7":"#f59e0b"}
for _, row in cstate_df.iterrows():
c0 = max(0.0, 100 - float(row.c1 or 0) - float(row.c2 or 0)
- float(row.c3 or 0) - float(row.c6 or 0) - float(row.c7 or 0))
cs_data = pd.DataFrame([
{"State":"C0","Residency%": c0},
{"State":"C1","Residency%": float(row.c1 or 0)},
{"State":"C2","Residency%": float(row.c2 or 0)},
{"State":"C3","Residency%": float(row.c3 or 0)},
{"State":"C6","Residency%": float(row.c6 or 0)},
{"State":"C7","Residency%": float(row.c7 or 0)},
])
st.markdown(f"**{row.provider} · {row.workflow_type}** — "
f"{float(row.pkg_w or 0):.2f}W · {int(row.samples):,} samples")
fig = px.bar(cs_data, x="Residency%", y="State", orientation="h",
color="State", color_discrete_map=CSTATE_COLORS)
fig.update_layout(**PL, height=160, showlegend=False)
st.plotly_chart(fig, use_container_width=True)
st.info("Cloud: mostly C6/C7 (deep sleep between API calls). "
"Local: forced C0 throughout inference loop.")
else:
st.info("No cpu_samples yet — run experiments to populate.")
st.divider()
if not runs.empty and "ipc" in runs.columns:
col1, col2 = st.columns(2)
with col1:
st.markdown("**IPC Distribution**")
_ri = runs.dropna(subset=["ipc"])
fig = px.histogram(_ri, x="ipc", color="workflow_type",
color_discrete_map=WF_COLORS,
nbins=20, barmode="overlay", opacity=.75,
labels={"ipc":"IPC"})
st.plotly_chart(fl(fig), use_container_width=True)
with col2:
st.markdown("**Cache Miss vs Energy**")
if "cache_miss_rate" in runs.columns and "energy_j" in runs.columns:
_rm = runs.dropna(subset=["cache_miss_rate","energy_j"]).copy()
_rm["cache_miss_pct"] = _rm["cache_miss_rate"] * 100
fig2 = px.scatter(_rm, x="cache_miss_pct", y="energy_j",
color="workflow_type", color_discrete_map=WF_COLORS,
log_y=True, hover_data=["run_id","provider"],
labels={"cache_miss_pct":"Cache Miss %","energy_j":"Energy J"})
st.plotly_chart(fl(fig2), use_container_width=True)
# ══════════════════════════════════════════════════════════════════════════════
# SCHEDULER
# ══════════════════════════════════════════════════════════════════════════════