Updated debloat and debloat.sh slightly

debloat does not set the bql limit at 10Mbit as low as debloat.sh now
does.

Still need to detect bridges

src/debloat

@@ -114,7 +114,7 @@ is helpful.
 This started out life as a shell script to exercise qfq,
 Now it does a lot more than that and is getting crufty. 
 
-SFQ is now the default. SFQ itself has been improved significantly
+FQ_CODEL is now the default. SFQ has been improved significantly
 in Linux 3.3 (eliminating a head of line problem), and in this case
 no new TC utility is required. Also a bug in red was fixed, and no
 new tc utility is required there either. So if you were using either
@@ -135,20 +135,20 @@ and with GSO on, at 100Mbit, I have seen latency spikes of up to 70ms.
 
    (Not recently tested, however)
 
-A per queue limit of 2-3 large packets appears to be the best
+A per queue limit of 2 large packets appears to be the best
 compromise at 100Mbit and below. So typically I hammer down BQL to
-4.5k or 3k at < 100Mbit, and turn GSO/TSO off, and as a result see
+3k at < 100Mbit, and turn GSO/TSO off, and as a result see
 ping against load latencies in the 1 to 2ms range, which is about
 what you would expect. I have tried 1500 bytes, which limited the top
-end performance to about 84Mbit.
+end performance to about 84Mbit. At 10Mbit, 1514 works on most OSes.
 
 For comparison, you will see PFIFO_FAST doing 130+ms, pre BQL, no
 SFQ at 100Mbit.
 
 * A BQL enabled ethernet device driver is helpful
 
 But there is currently no good way to detect if you have one at run
-time. 6-7 of the most major drivers have been convered to BQL, more
+time. 10 of the most major drivers have been convered to BQL, more
 remain.
 
 * Wireless still has problems
@@ -199,10 +199,12 @@ I have tried pfifo_drop_head, SFB, and RED here.  All had bugs until
 pfifo_drop_head generates interesting results.
 
 The very new combination of REDSFQ which compensates for both bytes
-and packets is very interesting, as it combines everything we have
+and packets was very interesting, as it combines everything we have
 learned in the past year into one single qdisc which can be brought up
 as a shaper in three lines of code.
 
+FQ_Codel is better.
+
 In other news:
 
 I have not tried the new 'adaptive red' implementation as a stand
@@ -1619,7 +1621,7 @@ local function ethernet_fq_codel(queues)
 end
 
 local function ethernet_fq_codel_ll(queues)
-	 qa("handle %x root fq_codel quantum 1000 ",10)
+	 qa("handle %x root fq_codel limit 1000 quantum 1000 ",10)
 end
 
 local function ethernet_codel(queues)

src/debloat.sh

@@ -8,8 +8,12 @@
 
 LL=1 # go for lowest latency
 ECN=1 # enable ECN
-BQLLIMIT=3000 # at speeds below 100Mbit, 2 big packets is enough
+BQLLIMIT100=3000 # at speeds below 100Mbit, 2 big packets is enough
+BQLLIMIT10=1514 # at speeds below 10Mbit, 1 big packet is enough. 
+		# Actually it would be nice to go to just one packet
 QDISC=fq_codel # There are multiple variants of fq_codel in testing
+FQ_LIMIT="" # the default 10000 packet limit mucks with slow start at speeds 
+            # at 1Gbit and below. Somewhat arbitrary figures selected.
 
 [ -z "$IFACE" ] && echo error: $0 expects IFACE parameter in environment && exit 1
 [ -z `which ethtool` ] && echo error: ethtool is required && exit 1
@@ -19,15 +23,19 @@ QDISC=fq_codel # There are multiple variants of fq_codel in testing
 # BUGS - need to detect bridges. 
 #      - Need filter to distribute across mq ethernet devices
 #      - needs an "undebloat" script for ifdown to restore BQL autotuning
-#      - should use a lower $QDISC limit at wifi and <10Gbit
 
 S=/sys/class/net
 FQ_OPTS=""
-#FQ_OPTS="FLOWS 2048 TARGET 5ms LIMIT 1000"
+#FQ_OPTS="FLOWS 2048 TARGET 5ms"
 
 [ $LL -eq 1 ] && FQ_OPTS="$FQ_OPTS quantum 500"
 [ $ECN -eq 1 ] && FQ_OPTS="$FQ_OPTS ecn"
 
+FLOW_KEYS="src,dst,proto,proto-src,proto-dst"
+# For 5-tuple (flow) fairness when the same device is performing NAT
+#FLOW_KEYS="nfct-src,nfct-dst,nfct-proto,nfct-proto-src,nfct-proto-dst"
+
+
 # Offloads are evil in the quest for low latency
 # And ethtool will abort if you attempt to turn off a
 # nonexistent offload.
@@ -66,20 +74,28 @@ mq() {
 		tc qdisc add dev $IFACE parent 1:$(printf "%x" $I) $QDISC $FQ_OPTS
 		I=`expr $I + 1`
 	done
+	I=`expr $I - 1`
+	tc filter add dev $IFACE prio 1 protocol ip parent 1: handle 100 \
+		flow hash keys ${FLOW_KEYS} divisor $I baseclass 1:1
 }
 
 fq_codel() {
-	tc qdisc add dev $IFACE root $QDISC $FQ_OPTS
+	tc qdisc add dev $IFACE root $QDISC $FQ_OPTS $FQ_LIMIT
 }
 
 fix_speed() {
 local SPEED=`cat $S/$IFACE/speed` 2> /dev/null
 if [ -n "$SPEED" ]
 then
+	[ "$SPEED" = 4294967295 ] && echo "no ethernet speed selected. debloat estimate will be WRONG"
+	[ "$SPEED" -lt 1001 ] && FQ_LIMIT=1200
 	if [ "$SPEED" -lt 101 ]
 	then
 	[ $LL -eq 1 ] && et # for lowest latency disable offloads
-		for I in /sys/class/net/$IFACE/queues/tx-*/byte_queue_limits/limit_max
+	BQLLIMIT=$BQLLIMIT100
+	FQ_LIMIT="limit 800"
+	[ "$SPEED" -lt 11 ] && BQLLIMIT=$BQLLIMIT10 && FQ_LIMIT="limit 400"
+	for I in /sys/class/net/$IFACE/queues/tx-*/byte_queue_limits/limit_max
 		do
 			echo $BQLLIMIT > $I
 		done