Skip to content

Commit

Permalink
unify codel params-clamp target to set values-do not adjust to appare…
Browse files Browse the repository at this point in the history 
…nt mtu.

We have a long standing "knowledge" that when target > than what the mtu
can serve at the available bandwidth...

		"bad things" happen to codel'd packets.

What we don't have, is extensive characterization of what that actually
looks like on a variety of tests.

So let's go get that data. THEN we can start talking about
mtu * 1.5 or any other value whatsoever, or gasp! even an
algorithm to calculate a smarter target based on the number of flows.
  • Loading branch information
Dave Taht committed Nov 25, 2015
1 parent b9cbdee commit 4b9e603
Showing 1 changed file with 15 additions and 31 deletions.
46 changes: 15 additions & 31 deletions sch_cake.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -133,7 +133,6 @@ struct cake_tin_data {
u16 bulk_flow_count;

u32 drop_overlimit;
struct codel_params cparams;

struct list_head new_flows; /* list of new flows */
struct list_head old_flows; /* list of old flows */
Expand All @@ -157,6 +156,7 @@ struct cake_tin_data {

struct cake_sched_data {
struct cake_tin_data *tins;
struct codel_params cparams;
u16 tin_cnt;
u8 tin_mode;
u8 flow_mode;
Expand Down Expand Up @@ -639,7 +639,7 @@ static struct sk_buff *cake_dequeue(struct Qdisc *sch)
prev_drop_count = flow->cvars.drop_count;
prev_ecn_mark = flow->cvars.ecn_mark;

skb = codel_dequeue(sch, &flow->cvars, &b->cparams, now,
skb = codel_dequeue(sch, &flow->cvars, &q->cparams, now,
q->buffer_used >
(q->buffer_limit >> 2) + (q->buffer_limit >> 1));

Expand Down Expand Up @@ -704,17 +704,14 @@ static const struct nla_policy cake_policy[TCA_CAKE_MAX + 1] = {
[TCA_CAKE_WASH] = { .type = NLA_U32 },
};

static void cake_set_rate(struct cake_tin_data *b, u64 rate, u32 mtu,
codel_time_t ns_target, codel_time_t rtt_est_ns)
static void cake_set_rate(struct cake_tin_data *b, u64 rate)
{
/* convert byte-rate into time-per-byte
* so it will always unwedge in reasonable time.
*/
static const u64 MIN_RATE = 64;
u64 rate_ns = 0;
u8 rate_shft = 0;
codel_time_t byte_target_ns;
u32 byte_target = mtu + (mtu >> 1);

if (rate) {
rate_shft = 32;
Expand All @@ -730,12 +727,6 @@ static void cake_set_rate(struct cake_tin_data *b, u64 rate, u32 mtu,
b->tin_rate_ns = rate_ns;
b->tin_rate_shft = rate_shft;

byte_target_ns = (byte_target * rate_ns) >> rate_shft;

b->cparams.target = max(byte_target_ns, ns_target);
b->cparams.interval = max(rtt_est_ns +
b->cparams.target - ns_target,
b->cparams.target * 8);
b->quantum = max(min(rate >> 12, 1514ULL), 300ULL);
}

Expand All @@ -744,15 +735,14 @@ static void cake_config_besteffort(struct Qdisc *sch)
struct cake_sched_data *q = qdisc_priv(sch);
struct cake_tin_data *b = &q->tins[0];
u64 rate = q->rate_bps;
u32 mtu = psched_mtu(qdisc_dev(sch));
u32 i;

q->tin_cnt = 1;

for (i = 0; i < 64; i++)
q->tin_index[i] = 0;

cake_set_rate(b, rate, mtu, US2TIME(q->target), US2TIME(q->interval));
cake_set_rate(b, rate);
b->tin_quantum_band = 65535;
b->tin_quantum_prio = 65535;
}
Expand All @@ -762,7 +752,6 @@ static void cake_config_precedence(struct Qdisc *sch)
/* convert high-level (user visible) parameters into internal format */
struct cake_sched_data *q = qdisc_priv(sch);
u64 rate = q->rate_bps;
u32 mtu = psched_mtu(qdisc_dev(sch));
u32 quantum1 = 256;
u32 quantum2 = 256;
u32 i;
Expand All @@ -775,8 +764,7 @@ static void cake_config_precedence(struct Qdisc *sch)
for (i = 0; i < q->tin_cnt; i++) {
struct cake_tin_data *b = &q->tins[i];

cake_set_rate(b, rate, mtu, US2TIME(q->target),
US2TIME(q->interval));
cake_set_rate(b, rate);

b->tin_quantum_prio = max_t(u16, 1U, quantum1);
b->tin_quantum_band = max_t(u16, 1U, quantum2);
Expand Down Expand Up @@ -854,7 +842,6 @@ static void cake_config_diffserv8(struct Qdisc *sch)

struct cake_sched_data *q = qdisc_priv(sch);
u64 rate = q->rate_bps;
u32 mtu = psched_mtu(qdisc_dev(sch));
u32 quantum1 = 256;
u32 quantum2 = 256;
u32 i;
Expand Down Expand Up @@ -889,8 +876,7 @@ static void cake_config_diffserv8(struct Qdisc *sch)
for (i = 0; i < q->tin_cnt; i++) {
struct cake_tin_data *b = &q->tins[i];

cake_set_rate(b, rate, mtu, US2TIME(q->target),
US2TIME(q->interval));
cake_set_rate(b, rate);

b->tin_quantum_prio = max_t(u16, 1U, quantum1);
b->tin_quantum_band = max_t(u16, 1U, quantum2);
Expand Down Expand Up @@ -921,7 +907,6 @@ static void cake_config_diffserv4(struct Qdisc *sch)

struct cake_sched_data *q = qdisc_priv(sch);
u64 rate = q->rate_bps;
u32 mtu = psched_mtu(qdisc_dev(sch));
u32 quantum = 256;
u32 i;

Expand Down Expand Up @@ -952,14 +937,10 @@ static void cake_config_diffserv4(struct Qdisc *sch)
}

/* class characteristics */
cake_set_rate(&q->tins[0], rate, mtu,
US2TIME(q->target), US2TIME(q->interval));
cake_set_rate(&q->tins[1], rate - (rate >> 4), mtu,
US2TIME(q->target), US2TIME(q->interval));
cake_set_rate(&q->tins[2], rate - (rate >> 2), mtu,
US2TIME(q->target), US2TIME(q->interval));
cake_set_rate(&q->tins[3], rate >> 2, mtu,
US2TIME(q->target), US2TIME(q->interval));
cake_set_rate(&q->tins[0], rate);
cake_set_rate(&q->tins[1], rate - (rate >> 4));
cake_set_rate(&q->tins[2], rate - (rate >> 2));
cake_set_rate(&q->tins[3], rate >> 2);

/* priority weights */
q->tins[0].tin_quantum_prio = quantum >> 4;
Expand Down Expand Up @@ -1017,6 +998,9 @@ static void cake_reconfigure(struct Qdisc *sch)
q->buffer_limit = ~0;
}

q->cparams.target = max_t(u64,US2TIME(q->target),0);
q->cparams.interval = US2TIME(q->interval);

if (q->rate_bps)
sch->flags &= ~TCQ_F_CAN_BYPASS;
else
Expand Down Expand Up @@ -1262,9 +1246,9 @@ static int cake_dump_stats(struct Qdisc *sch, struct gnet_dump *d)
struct cake_tin_data *b = &q->tins[i];

st->threshold_rate[i] = b->tin_rate_bps;
st->target_us[i] = codel_time_to_us(b->cparams.target);
st->target_us[i] = codel_time_to_us(q->cparams.target);
st->interval_us[i] =
codel_time_to_us(b->cparams.interval);
codel_time_to_us(q->cparams.interval);

/* TODO FIXME: add missing aspects of these composite stats */
st->sent[i].packets = b->packets;
Expand Down

0 comments on commit 4b9e603

Please sign in to comment.