Add infrastructure for measuring hugepage bloat

kernel/src/arch/x86/pmu.rs

@@ -13,13 +13,13 @@ use ostd::{
     },
     path,
 };
+use serde::Serialize;
 use snafu::Whatever;
 
 use crate::util::timer::TimerServer;
 
 /// Data TLB Misses instance struct
-#[derive(Debug, Clone, Copy)]
-#[expect(dead_code)]
+#[derive(Debug, Clone, Copy, Serialize)]
 struct DtlbMisses {
     timestamp: Instant,
     miss_l1_tlb: u64,

kernel/src/thread/exception.rs

@@ -4,6 +4,7 @@
 
 use aster_rights::Full;
 use ostd::cpu::context::{CpuExceptionInfo, UserContext};
+use serde::Serialize;
 
 use crate::{
     current_userspace,
@@ -17,7 +18,7 @@ use crate::{
 /// `From<CpuExceptionInfo>` should be implemented for this struct.
 /// If `CpuExceptionInfo` is a page fault, `try_from` should return `Ok(PageFaultInfo)`,
 /// or `Err(())` (no error information) otherwise.
-#[derive(Clone, Copy)]
+#[derive(Clone, Copy, Serialize)]
 pub struct PageFaultInfo {
     /// The virtual address where a page fault occurred.
     pub address: Vaddr,

kernel/src/vm/hugepaged.rs

@@ -11,7 +11,7 @@ use ostd::orpc::{
 };
 use snafu::Whatever;
 
-use super::{promote_hugepages, vmar::PageFaultOQueueMessage};
+use super::{promote_hugepages, vmar::oqueues::PageFaultOQueueMessage};
 use crate::{process::Process, util::timer::TimerServer, vm::vmar};
 
 #[orpc_trait]

kernel/src/vm/mod.rs

@@ -27,14 +27,16 @@ use ostd::{
     mm::{FrameAllocOptions, PageFlags, PageProperty, PagingConsts, UFrame, UntypedMem, page_size},
     task::disable_preempt,
 };
-use vmar::{PageFaultOQueueMessage, RssType};
+use vmar::RssType;
 
 use crate::{
     INITPROC, Vec,
     process::{PauseProcGuard, Process},
 };
 #[cfg(not(baseline_asterinas))]
 pub mod hugepaged;
+#[cfg(not(baseline_asterinas))]
+use vmar::oqueues::PageFaultOQueueMessage;
 
 pub mod page_fault_handler;
 pub mod perms;
@@ -108,6 +110,7 @@ pub fn num_anon_hugepages() -> i32 {
     count
 }
 
+#[cfg(not(baseline_asterinas))]
 fn promote_hugepages(
     proc: &Arc<Process>,
     fault_hint: Option<PageFaultOQueueMessage>,

kernel/src/vm/perms.rs

@@ -3,9 +3,11 @@
 use aster_rights::Rights;
 use bitflags::bitflags;
 use ostd::mm::PageFlags;
+use serde::Serialize;
 
 bitflags! {
     /// The memory access permissions of memory mappings.
+    #[derive(Serialize)]
     pub struct VmPerms: u32 {
         /// Readable.
         const READ    = 1 << 0;

kernel/src/vm/vmar/mod.rs

@@ -188,35 +188,35 @@ impl<R> Vmar<R> {
     }
 }
 
-// TODO(aneesh) can this just be PageFaultInfo directly?
-/// Notification message to inform policies about a page fault
-#[derive(Clone, Copy)]
-pub struct PageFaultOQueueMessage {
-    /// Opaque identifier for which vm_space the fault corresponds to
-    pub vm_space_id: u64,
-    /// The fault information provided to the page fault handler
-    pub fault_info: PageFaultInfo,
-}
-
 #[cfg(not(baseline_asterinas))]
 pub mod oqueues {
     use alloc::sync::Arc;
     use core::{sync::atomic::AtomicUsize, time::Duration};
 
     use ostd::orpc::legacy_oqueue::ringbuffer::MPMCOQueue;
+    use serde::Serialize;
     use spin::Once;
 
-    use super::PageFaultOQueueMessage;
     use crate::{Clock, time::clocks::MonotonicRawClock};
 
+    // TODO(aneesh) can this just be PageFaultInfo directly?
+    /// Notification message to inform policies about a page fault
+    #[derive(Clone, Copy, Serialize)]
+    pub struct PageFaultOQueueMessage {
+        /// Opaque identifier for which vm_space the fault corresponds to
+        pub vm_space_id: u64,
+        /// The fault information provided to the page fault handler
+        pub fault_info: super::PageFaultInfo,
+    }
+
     // TODO(aneesh): Move this somewhere more generic
-    #[derive(Clone, Copy)]
-    pub struct ObservableEvent<T> {
+    #[derive(Clone, Copy, Serialize)]
+    pub struct ObservableEvent<T: Serialize> {
         pub event: T,
         pub timestamp: Duration,
     }
 
-    impl<T> ObservableEvent<T> {
+    impl<T: Serialize> ObservableEvent<T> {
         pub fn new(event: T) -> Self {
             Self {
                 event,
@@ -245,9 +245,9 @@ pub mod oqueues {
 pub fn init() {
     #[cfg(not(baseline_asterinas))]
     {
-        // Only support a single strong observer for now - hugepaged.
-        oqueues::PAGE_FAULT_OQUEUE.call_once(|| MPMCOQueue::new(64, 1));
-        oqueues::RSS_DELTA_OQUEUE.call_once(|| MPMCOQueue::new(64, 1));
+        // Supports two strong observers for now - hugepaged and data collection.
+        oqueues::PAGE_FAULT_OQUEUE.call_once(|| MPMCOQueue::new(1024, 2));
+        oqueues::RSS_DELTA_OQUEUE.call_once(|| MPMCOQueue::new(1024, 2));
     }
 }
 
@@ -266,7 +266,8 @@ pub(super) struct Vmar_ {
 
     /// OQueue Producer to notify policies about page fault events
     #[cfg(not(baseline_asterinas))]
-    page_fault_oqueue_producer: OQueueRef<oqueues::ObservableEvent<PageFaultOQueueMessage>>,
+    page_fault_oqueue_producer:
+        OQueueRef<oqueues::ObservableEvent<oqueues::PageFaultOQueueMessage>>,
 }
 
 struct VmarInner {
@@ -661,10 +662,12 @@ impl Vmar_ {
             #[cfg(not(baseline_asterinas))]
             if res.is_ok() {
                 self.page_fault_oqueue_producer
-                    .produce(oqueues::ObservableEvent::new(PageFaultOQueueMessage {
-                        vm_space_id: self.vm_space.id(),
-                        fault_info: *page_fault_info,
-                    }))?;
+                    .produce(oqueues::ObservableEvent::new(
+                        oqueues::PageFaultOQueueMessage {
+                            vm_space_id: self.vm_space.id(),
+                            fault_info: *page_fault_info,
+                        },
+                    ))?;
             }
             return res;
         }

test/benchmark/common/prepare_host.sh

@@ -68,7 +68,8 @@ prepare_ycsb() {
     fi
 
     if [ ! -d "$YCSB_PATH" ]; then
-      git clone https://github.com/brianfrankcooper/YCSB --depth=1 $YCSB_PATH
+      # Use custom fork of YCSB with delete support
+      git clone https://github.com/tewaro/YCSB.git -b tewaro/quickfix-coreworkload-deletes-master --depth=1 $YCSB_PATH
 
       # Build
       pushd $YCSB_PATH

test/benchmark/redis/ycsb/host.sh

@@ -7,6 +7,9 @@ set -e
 # Function to stop the guest VM
 stop_guest() {
     echo "Stopping guest VM..."
+    # The message doesn't matter here, any data will make the vm exit
+    echo stop | nc 10.0.2.15 5201
+    sleep 15
     # `-r` means if there's no qemu, the kill won't be executed
     pgrep qemu | xargs -r kill
 }
@@ -22,7 +25,37 @@ echo "Running YCSB bench connected to $GUEST_SERVER_IP_ADDRESS"
 export JAVA_HOME=$(realpath ".cache/jdk-25.0.2")
 
 cd $YCSB_PATH/
-./bin/ycsb load redis -p redis.host="$GUEST_SERVER_IP_ADDRESS" -p redis.port="6379" -P ./workloads/workloada
-./bin/ycsb run redis -p redis.host="$GUEST_SERVER_IP_ADDRESS" -p redis.port="6379" -P ./workloads/workloada
+
+FIELDLENGTH=4096
+
+
+
+./bin/ycsb load redis -p redis.host="$GUEST_SERVER_IP_ADDRESS" -p redis.port="6379" -P ./workloads/workloada \
+  -p recordcount=4096 \
+  -p fieldcount=1 \
+  -p fieldlength=$FIELDLENGTH \
+  -p minfieldlength=4096 \
+  -p insertstart=0 \
+  -p fieldlengthdistribution=uniform
+
+# Run many times to stress memory allocation
+for _ in $(seq 1 4096); do
+  ./bin/ycsb run redis -p redis.host="$GUEST_SERVER_IP_ADDRESS" -p redis.port="6379" -P ./workloads/workloada \
+    -p operationcount=4096 \
+    -p recordcount=4096 \
+    -p workload=site.ycsb.workloads.CoreWorkload \
+    -p readproportion=0.05 \
+    -p updateproportion=0.00 \
+    -p scanproportion=0.00 \
+    -p insertproportion=0.00 \
+    -p readmodifywriteproportion=0.00 \
+    -p scanproportion=0.00 \
+    -p deleteproportion=0.95 \
+    -p threadcount=16 \
+    -p fieldcount=1 \
+    -p fieldlength=$FIELDLENGTH \
+    -p minfieldlength=4096 \
+    -p fieldlengthdistribution=uniform
+done
 
 # The trap will automatically stop the guest VM when the script exits

test/benchmark/redis/ycsb/run.sh

@@ -2,7 +2,8 @@
 
 # SPDX-License-Identifier: MPL-2.0
 
-set -e
-
 echo "Running redis server"
-/usr/local/redis/bin/redis-server /benchmark/redis/ycsb/ycsb.conf
+/usr/local/redis/bin/redis-server /benchmark/redis/ycsb/ycsb.conf &
+echo "waiting for stop message"
+/bin/nc -l -s 10.0.2.15 -p 5201 -e true
+echo "done"

tools/docker/Dockerfile

@@ -24,6 +24,9 @@ RUN apt update && apt-get install -y --no-install-recommends \
     unzip \
     zip
 
+# Install netcat for clean shutdown of benchmarks
+# see test/benchmark/redis/ycsb/host.sh for example usage
+RUN apt-get install -y --no-install -recommends netcat-openbsd
 
 #= Install linux-modules for vdso and linux config =================================
 # TODO: This is very wasteful and maybe fragile. We should have our own custom vDSO.