diff --git a/crates/buffet/src/net/net_uring.rs b/crates/buffet/src/net/net_uring.rs
index aa7e8acf..3f54e7c3 100644
--- a/crates/buffet/src/net/net_uring.rs
+++ b/crates/buffet/src/net/net_uring.rs
@@ -175,34 +175,37 @@ impl WriteOwned for TcpWriteHalf {
 
     // TODO: implement writev
 
-    // async fn writev_owned(&mut self, list: &PieceList) -> std::io::Result<usize>
-    // {     let mut iovecs: Vec<iovec> = vec![];
-    //     for piece in list.pieces.iter() {
-    //         let iov = iovec {
-    //             iov_base: piece.as_ref().as_ptr() as *mut libc::c_void,
-    //             iov_len: piece.len(),
-    //         };
-    //         iovecs.push(iov);
-    //     }
-    //     let iovecs = iovecs.into_boxed_slice();
-    //     let iov_cnt = iovecs.len();
-    //     // FIXME: don't leak, duh
-    //     let iovecs = Box::leak(iovecs);
-
-    //     let sqe = Writev::new(
-    //         io_uring::types::Fd(self.0.fd),
-    //         iovecs.as_ptr() as *const _,
-    //         iov_cnt as u32,
-    //     )
-    //     .build();
-
-    //     let cqe = get_ring().push(sqe).await;
-    //     let ret = match cqe.error_for_errno() {
-    //         Ok(ret) => ret,
-    //         Err(e) => return Err(std::io::Error::from(e)),
-    //     };
-    //     Ok(ret as usize)
-    // }
+    async fn writev_owned(&mut self, list: &crate::PieceList) -> std::io::Result<usize> {
+        use libc::iovec;
+        use io_uring::opcode::Writev;
+
+        let mut iovecs: Vec<iovec> = vec![];
+        for piece in list.pieces.iter() {
+            let iov = iovec {
+                iov_base: piece.as_ref().as_ptr() as *mut libc::c_void,
+                iov_len: piece.len(),
+            };
+            iovecs.push(iov);
+        }
+        let iovecs = iovecs.into_boxed_slice();
+        let iov_cnt = iovecs.len();
+        // FIXME: don't leak, duh
+        let iovecs = Box::leak(iovecs);
+
+        let sqe = Writev::new(
+            io_uring::types::Fd(self.0.fd),
+            iovecs.as_ptr() as *const _,
+            iov_cnt as u32,
+        )
+        .build();
+
+        let cqe = get_ring().push(sqe).await;
+        let ret = match cqe.error_for_errno() {
+            Ok(ret) => ret,
+            Err(e) => return Err(std::io::Error::from(e)),
+        };
+        Ok(ret as usize)
+    }
 
     async fn shutdown(&mut self) -> std::io::Result<()> {
         tracing::debug!("requesting shutdown");
diff --git a/crates/loona/src/h2/server.rs b/crates/loona/src/h2/server.rs
index 66a1a691..d764d9e1 100644
--- a/crates/loona/src/h2/server.rs
+++ b/crates/loona/src/h2/server.rs
@@ -187,7 +187,6 @@ where
         let mut goaway_err: Option<H2ConnectionError> = None;
 
         {
-            // read frames and send them into an mpsc buffer of size 1
             let (tx, rx) = mpsc::channel::<(Frame, Roll)>(32);
 
             // store max frame size setting as an atomic so we can share it across tasks
diff --git a/scripts/perfstat.sh b/scripts/perfstat.sh
index 02a40850..779b90c6 100755
--- a/scripts/perfstat.sh
+++ b/scripts/perfstat.sh
@@ -15,7 +15,7 @@ trap 'kill -TERM -$$' EXIT
 mkdir -p /tmp/loona-perfstat
 
 # Kill older processes
-for pidfile in /tmp/loona-perfstat/{hyper,loona}.PID; do
+for pidfile in /tmp/loona-perfstat/*.PID; do
     if [ -f "$pidfile" ]; then
         pid=$(cat "$pidfile")
         if [ "$pid" != "$$" ]; then
@@ -97,11 +97,34 @@ CONNS="${CONNS:-40}"
 STREAMS="${STREAMS:-8}"
 NUM_REQUESTS="${NUM_REQUESTS:-100}"
 
+# Set MODE to 'stat' if not specified
+MODE=${MODE:-stat}
+
+if [[ "$MODE" == "record" ]]; then
+    PERF_CMD="perf record -F 99 -e $PERF_EVENTS -p"
+elif [[ "$MODE" == "stat" ]]; then
+    PERF_CMD="perf stat -e $PERF_EVENTS -p"
+else
+    echo "Error: Unknown MODE '$MODE'"
+    exit 1
+fi
+
 echo -e "\033[1;34m📊 Benchmark parameters: RPS=$RPS, CONNS=$CONNS, STREAMS=$STREAMS, NUM_REQUESTS=$NUM_REQUESTS, ENDPOINT=$ENDPOINT\033[0m"
 
 for server in "${!servers[@]}"; do
     read -r PID ADDR <<< "${servers[$server]}"
     echo -e "\033[1;36mLoona Git SHA: $(cd ~/bearcove/loona && git rev-parse --short HEAD)\033[0m"
     echo -e "\033[1;33m🚀 Benchmarking \033[1;32m$(cat /proc/$PID/cmdline | tr '\0' ' ')\033[0m"
-    perf stat -e "$PERF_EVENTS" -p "$PID" -- ssh brat "$H2LOAD" "${H2LOAD_ARGS[@]}" --rps "$RPS" -c "$CONNS" -m "$STREAMS" -n "$NUM_REQUESTS" "${ADDR}${ENDPOINT}"
+    remote_command=("$H2LOAD" "${H2LOAD_ARGS[@]}" --rps "$RPS" -c "$CONNS" -m "$STREAMS" -n "$NUM_REQUESTS" "${ADDR}${ENDPOINT}")
+
+    if [[ "$MODE" == "record" ]]; then
+        samply record -p "$PID" &
+        SAMPLY_PID=$!
+        echo $SAMPLY_PID > /tmp/loona-perfstat/samply.PID
+        ssh brat "${remote_command[@]}"
+        kill -INT $SAMPLY_PID
+        wait $SAMPLY_PID
+    else
+        perf stat -e "$PERF_EVENTS" -p "$PID" -- ssh brat "${remote_command[@]}"
+    fi
 done
diff --git a/scripts/syscalls b/scripts/syscalls
index 573bb944..81e6fc06 100644
--- a/scripts/syscalls
+++ b/scripts/syscalls
@@ -1,5 +1,4 @@
-syscalls:sys_enter_accept*
-syscalls:sys_enter_write*
-syscalls:sys_enter_brk
+syscalls:sys_enter_write
+syscalls:sys_enter_writev
 syscalls:sys_enter_epoll_wait
 syscalls:sys_enter_io_uring_enter