diff --git a/tools/biosnoop.py b/tools/biosnoop.py
index 37ee3f9..21f74c5 100755
--- a/tools/biosnoop.py
+++ b/tools/biosnoop.py
@@ -11,9 +11,13 @@
 # Licensed under the Apache License, Version 2.0 (the "License")
 #
 # 16-Sep-2015   Brendan Gregg   Created this.
+# 11-Feb-2016   Allan McAleavy  updated for BPF_PERF_OUTPUT
+
 
 from __future__ import print_function
 from bcc import BPF
+import ctypes as ct
+import re
 
 # load BPF program
 b = BPF(text="""
@@ -25,8 +29,21 @@
     char name[TASK_COMM_LEN];
 };
 
+struct key_t {
+    u32 pid;
+    u64 rwflag;
+    u64 delta;
+    u64 sector;
+    u64 len;
+    u64 ts;
+    char disk_name[DISK_NAME_LEN];
+    char name[TASK_COMM_LEN];
+};
+
+
 BPF_HASH(start, struct request *);
 BPF_HASH(infobyreq, struct request *, struct val_t);
+BPF_PERF_OUTPUT(events);
 
 // cache PID and comm by-req
 int trace_pid_start(struct pt_regs *ctx, struct request *req)
@@ -37,7 +54,6 @@
         val.pid = bpf_get_current_pid_tgid();
         infobyreq.update(&req, &val);
     }
-
     return 0;
 }
 
@@ -58,6 +74,8 @@
     u64 *tsp, delta;
     u32 *pidp = 0;
     struct val_t *valp;
+    struct key_t key ={};
+    u64 ts;
 
     // fetch timestamp and calculate delta
     tsp = start.lookup(&req);
@@ -65,30 +83,29 @@
         // missed tracing issue
         return 0;
     }
-    delta = bpf_ktime_get_ns() - *tsp;
+    ts =  bpf_ktime_get_ns();
+    key.delta = ts - *tsp;
+    key.ts = ts / 1000;
 
-    //
-    // Fetch and output issuing pid and comm.
-    // As bpf_trace_prink() is limited to a maximum of 1 string and 2
-    // integers, we'll use more than one to output the data.
-    //
     valp = infobyreq.lookup(&req);
     if (valp == 0) {
-        bpf_trace_printk("0 0 ? %d\\n", req->__data_len);
+        key.len = req->__data_len;
+        strcpy(key.name,"?");
     } else {
-        bpf_trace_printk("0 %d %s %d\\n", valp->pid, valp->name,
-            req->__data_len);
+        key.pid = valp->pid;
+        key.len = req->__data_len;
+        key.sector = req->__sector;
+        bpf_probe_read(&key.name, sizeof(key.name), valp->name);
+        bpf_probe_read(&key.disk_name, sizeof(key.disk_name),
+                       req->rq_disk->disk_name);
     }
 
-    // output remaining details
     if (req->cmd_flags & REQ_WRITE) {
-        bpf_trace_printk("1 W %s %d %d ?\\n", req->rq_disk->disk_name,
-            req->__sector, delta / 1000);
+        key.rwflag=1;
     } else {
-        bpf_trace_printk("1 R %s %d %d ?\\n", req->rq_disk->disk_name,
-            req->__sector, delta / 1000);
+        key.rwflag=0;
     }
-
+    events.perf_submit(ctx,&key,sizeof(key));
     start.delete(&req);
     infobyreq.delete(&req);
 
@@ -101,29 +118,51 @@
 b.attach_kprobe(event="blk_account_io_completion",
     fn_name="trace_req_completion")
 
+
+TASK_COMM_LEN = 16  # linux/sched.h
+DISK_NAME_LEN = 32  # linux/genhd.h
+
+class Data(ct.Structure):
+    _fields_ = [
+        ("pid", ct.c_ulonglong),
+        ("rwflag", ct.c_ulonglong),
+        ("delta", ct.c_ulonglong),
+        ("sector", ct.c_ulonglong),
+        ("len", ct.c_ulonglong),
+        ("ts", ct.c_ulonglong),
+        ("disk_name", ct.c_char * DISK_NAME_LEN),
+        ("name", ct.c_char * TASK_COMM_LEN)
+    ]
 # header
 print("%-14s %-14s %-6s %-7s %-2s %-9s %-7s %7s" % ("TIME(s)", "COMM", "PID",
     "DISK", "T", "SECTOR", "BYTES", "LAT(ms)"))
 
+rwflg = ""
 start_ts = 0
-
-# format output
-while 1:
-    (task, pid, cpu, flags, ts, msg) = b.trace_fields()
-    args = msg.split(" ")
-
+prev_ts = 0
+delta = 0
+# process event
+def print_event(cpu, data, size):
+    event = ct.cast(data, ct.POINTER(Data)).contents
+    val = -1
+    global start_ts
+    global prev_ts
+    global delta
+    if event.rwflag == 1:
+        rwflg = "W"
+    if event.rwflag == 0:
+        rwflg = "R"
+    if not re.match('\?', event.name):
+        val = event.sector
     if start_ts == 0:
-        start_ts = ts
-
-    if args[0] == "0":
-        (real_pid, real_comm, bytes_s) = (args[1], args[2], args[3])
-        continue
-    else:
-        (type_s, disk_s, sector_s, us_s) = (args[1], args[2], args[3],
-            args[4])
-
-    ms = float(int(us_s, 10)) / 1000
-
+        prev_ts = start_ts
+    if start_ts == 1:
+        delta = float(delta) + (event.ts - prev_ts)
     print("%-14.9f %-14.14s %-6s %-7s %-2s %-9s %-7s %7.2f" % (
-        ts - start_ts, real_comm, real_pid, disk_s, type_s, sector_s,
-        bytes_s, ms))
+        delta / 1000000, event.name, event.pid, event.disk_name, rwflg, val,
+        event.len, float(event.delta) / 1000000))
+    prev_ts = event.ts
+    start_ts = 1
+b["events"].open_perf_buffer(print_event)
+while 1:
+    b.kprobe_poll()