Adopt 24 MiB mobile memory target

Author: bitprecipice

app/app/src/main/java/com/bringyour/network/DeviceManager.kt

@@ -22,7 +22,7 @@ class DeviceManager @Inject constructor(
         // provider share backing the client pair while providing is off. The
         // process-level Sdk.setMemoryLimit (MainApplication) sizes the
         // shared message pools and go soft limit separately.
-        const val DEVICE_MEMORY_TARGET_BYTE_COUNT = 20L * 1024 * 1024
+        const val DEVICE_MEMORY_TARGET_BYTE_COUNT = 24L * 1024 * 1024
     }
 
     private val deviceLock = Any()

app/app/src/main/java/com/bringyour/network/MainApplication.kt

@@ -40,7 +40,7 @@ class MainApplication : Application() {
         const val VPN_STATE_BURST_COALESCE_MILLIS = 20L
         // Match the iOS packet-tunnel process budget so Android physical runs
         // expose the same SDK pressure/failure boundary. DeviceManager already
-        // passes the matching iOS per-device target (20 MiB).
+        // passes the matching iOS per-device steady target (24 MiB).
         const val SDK_PROCESS_MEMORY_LIMIT_MIB = 32L
     }
 

app/app/src/test/java/com/bringyour/network/MobileMemoryPolicyTest.kt

@@ -0,0 +1,11 @@
+package com.bringyour.network
+
+import org.junit.Assert.assertEquals
+import org.junit.Test
+
+class MobileMemoryPolicyTest {
+    @Test
+    fun deviceUsesTwentyFourMiBSteadyTarget() {
+        assertEquals(24L * 1024 * 1024, DeviceManager.DEVICE_MEMORY_TARGET_BYTE_COUNT)
+    }
+}

app/scripts/PHYSICAL_LOWBAR.md

@@ -94,14 +94,15 @@ before/after values, aggregate mobile packet-pressure drops, primitive
 client/flow topology, platform transport-budget use, Android PSS, Java/native
 heaps, threads, descriptors, route state, and aggregate carrier counters. The
 steady streamline signal is `goRuntimeBytes`:
-five quiet connected minutes should have p50 and p95 at or below 20 MiB. Keep
+five quiet connected minutes after burst ownership drains should have p50 and
+p95 at or below 24 MiB. Keep
 the active peak and time-to-recover separate, and investigate every sample over
 the 28-MiB active diagnostic threshold. Neither threshold is Android whole-app
 PSS or an iOS Network Extension `phys_footprint` ceiling.
 
 Ordinary Android and Apple SDK libraries start the Go runtime with
-`memprofilerate=0`. Android and iOS also use the same `GOGC=10` pacing for the
-20-MiB campaign; a looser Android heap float is not a valid surrogate for the
+`memprofilerate=0`. Android and iOS also use the same `GOGC=25` pacing for the
+24-MiB campaign; a looser Android heap float is not a valid surrogate for the
 iOS Network Extension. A private diagnostic build can opt in before native
 runtime initialization with `-PurnetworkMemoryProfileRateBytes=65536`; the
 Gradle value is passed to both the AAR linker and the diagnostic app API.
@@ -115,10 +116,15 @@ mobile reclaimer waits for payload quiet and bounded outstanding ownership,
 then performs at most one pass per cooldown; use its deferred, below-target,
 cooldown, and before/after counters to distinguish policy from a leak.
 `packetPressureDropCount` is a cumulative overload counter, not a pool leak:
-the <=20-MiB mobile profile samples packet-root ownership every fourth ingress
+the <=24-MiB mobile profile samples packet-root ownership every fourth ingress
 call below pressure, rejects a complete native ingress batch at 512 or more
 roots, and samples every call until ownership drains. Rejected batch ownership
 is returned immediately; TCP retransmission/backpressure provides recovery.
+The accepted performance profile keeps the H3-safe 16-message sequence and
+16-packet/24-KiB group ceilings, fixes Auto quality/speed windows at 4/1, and
+retains a 512-KiB packet warm set after reclaim. Do not raise the aggregate
+gate or double the per-flow/group ceilings without repeating the explicit-H3
+28-MiB failure test; both experiments breached it on the physical surrogate.
 Server/default devices do not instantiate this gate. The same mobile profile
 retires inactive TCP flow state after three minutes so a closed browser burst
 does not preserve the desktop ten-minute graph throughout the five-minute