[Utilization] fix bugs (#6312)

remove the duration spike since it's not useful,
the gpu_mem_all should aggregated as max instead avg

Author: yangw-dev

torchci/components/utilization/components/UtilizationStatsTable.tsx

@@ -14,9 +14,6 @@ export default function UtilizationStatsTable({ data }: { data: any[] }) {
       spike_frequency: row.columns.find(
         (col: any) => col.type == StatType.SpikeFrequency
       )?.value,
-      spike_avg_interval: row.columns.find(
-        (col: any) => col.type == StatType.SpikeAvgInterval
-      )?.value,
     };
   });
   return (
@@ -93,10 +90,4 @@ const columns: GridColDef[] = [
     valueFormatter: valueFormatterSpike,
     minWidth: 200,
   },
-  {
-    field: "spike_avg_interval",
-    headerName: "avg spike interval (above 90%)",
-    valueFormatter: valueFormatterSeconds,
-    minWidth: 200,
-  },
 ];

torchci/components/utilization/helper.ts

@@ -105,13 +105,6 @@ export function getTimeSeriesStats(dps: TimeSeriesDataPoint[]) {
     unit: "%",
   };
   results.push(maxStat);
-  const avgSpikeInterval = findAvgSpikeIntervals(dps, p90Metric.value);
-  const spikeAvgInterval = {
-    type: StatType.SpikeAvgInterval,
-    value: Number((avgSpikeInterval / 1000).toFixed(2)),
-    unit: "%",
-  };
-  results.push(spikeAvgInterval);
   return results;
 }
 
@@ -121,17 +114,6 @@ export const findMaxValue = (data: number[]) => {
   return max;
 };
 
-const findAvgSpikeIntervals = (
-  timestamps: TimeSeriesDataPoint[],
-  threshold: number
-) => {
-  const spikeIntervals = findSpikeIntervals(timestamps, threshold);
-  if (spikeIntervals.length == 0) return -1; // No spikes
-  const avgSpike =
-    spikeIntervals.reduce((a, b) => a + b, 0) / spikeIntervals.length;
-  return avgSpike;
-};
-
 const findSpikeIntervals = (
   timestamps: TimeSeriesDataPoint[],
   threshold: number
@@ -306,7 +288,7 @@ function getAllGpusStats(stats: StatsInfo[]) {
       id: "gpu_mem_all",
       columns: [
         aggregateStats(gpuMems, StatType.Average, AgggregateMethod.Average),
-        aggregateStats(gpuMems, StatType.Max, AgggregateMethod.Average),
+        aggregateStats(gpuMems, StatType.Max, AgggregateMethod.Max),
         aggregateStats(
           gpuMems,
           StatType.SpikeFrequency,