vhost: Reduce TX used buffer signal for performance
Subsystems (4)
cover vhost: Reduce TX used buffer signal for performance
This patch will change vhost TX used buffer guest signaling from one by one to 3/4 ring size. I have tried different size, like 4, 16, 1/4 size, 1/2 size, and found that the large size is best for message size between 256 - 4K with netperf TCP_STREAM test, so 3/4 of the ring size is picked up for signaling.
Tested both UDP and TCP performance with guest 2vcpu. The 60 secs netperf run shows that guest to host performance for TCP.
TCP_STREAM
Message size Guest CPU(%) BW (Mb/s)
before:after before:after
256 57.84:58.42 1678.47:1908.75
512 68.68:60.21 1844.18:3387.33
1024 68.01:58.70 1945.14:3384.72
2048 65.36:54.25 2342.45:3799.31
4096 63.25:54.62 3307.11:4451.78
8192 59.57:57.89 6038.64:6694.04
UDP_STREAM
1024 49.64:26.69 1161.0:1687.6
2048 49.88:29.25 2326.8:2850.9
4096 49.59:29.15 3871.1:4880.3
8192 46.09:32.66 6822.9:7825.1
16K 42.90:34.96 11347.1:11767.4
For large message size, 60 secs run remains almost the same. I guess the signal might not play a big role in large message transmission.
Shirley
This patch changes vhost TX used buffer signal to guest from one by one to up to 3/4 of vring size. This change improves vhost TX message size from 256 to 8K performance for both bandwidth and CPU utilization without inducing any regression.
---
drivers/vhost/net.c | 19 ++++++++++++++++++- drivers/vhost/vhost.c | 31 +++++++++++++++++++++++++++++++ drivers/vhost/vhost.h | 3 +++ 3 files changed, 52 insertions(+), 1 deletions(-)
drivers/vhost/net.c +18 -1
@@ -198,7 +198,24 @@ static void handle_tx(struct vhost_net *net)if (err != len)pr_debug ("Truncated TX packet: " " len %d != %zd\n" , err, len); -vhost_add_used_and_signal (&net->dev, vq, head,0 ); +/* +* if no pending buffer size allocate, signal used buffer +* one by one, otherwise, signal used buffer when reaching +* 3/4 ring size to reduce CPU utilization. +*/ +if (unlikely (vq->pend)) +vhost_add_used_and_signal (&net->dev, vq, head,0 ); +else { + vq->pend[vq->num_pend].id = head; + vq->pend[vq->num_pend].len =0 ; + ++vq->num_pend; +if (vq->num_pend == (vq->num - (vq->num >>2 ))) { +vhost_add_used_and_signal_n (&net->dev, vq, + vq->pend, + vq->num_pend); + vq->num_pend =0 ; + } + } total_len += len;if (unlikely (total_len >=VHOST_NET_WEIGHT )) {vhost_poll_queue (&vq->poll);
drivers/vhost/vhost.c +31 -0
@@ -170,6 +170,16 @@ static void vhost_vq_reset(struct vhost_dev *dev, vq->call_ctx =NULL ; vq->call =NULL ; vq->log_ctx =NULL ; +/* signal pending used buffers */ +if (vq->pend) { +if (vq->num_pend !=0 ) { +vhost_add_used_and_signal_n (dev, vq, vq->pend, + vq->num_pend); + vq->num_pend =0 ; + } +kfree (vq->pend); + } + vq->pend =NULL ; }static int vhost_worker (void *data)
@@ -273,7 +283,13 @@ long vhost_dev_init(struct vhost_dev *dev, dev->vqs[i].heads =NULL ; dev->vqs[i].dev = dev;mutex_init (&dev->vqs[i].mutex); + dev->vqs[i].num_pend =0 ; + dev->vqs[i].pend =NULL ;vhost_vq_reset (dev, dev->vqs + i); +/* signal 3/4 of ring size used buffers */ + dev->vqs[i].pend =kmalloc ((dev->vqs[i].num - + (dev->vqs[i].num >>2 )) * +sizeof *vq->peed,GFP_KERNEL );if (dev->vqs[i].handle_kick)vhost_poll_init (&dev->vqs[i].poll, dev->vqs[i].handle_kick,POLLIN , dev);
@@ -599,6 +615,21 @@ static long vhost_set_vring(struct vhost_dev *d, int ioctl, void __user *argp) r = -EINVAL ;break ; } +if (vq->num != s.num) { +/* signal used buffers first */ +if (vq->pend) { +if (vq->num_pend !=0 ) { +vhost_add_used_and_signal_n (vq->dev, vq, + vq->pend, + vq->num_pend); + vq->num_pend =0 ; + } +kfree (vq->pend); + } +/* realloc pending used buffers size */ + vq->pend =kmalloc ((s.num - (s.num >>2 )) * +sizeof *vq->pend,GFP_KERNEL ); + } vq->num = s.num;break ;case VHOST_SET_VRING_BASE :
drivers/vhost/vhost.h +3 -0
@@ -108,6 +108,9 @@ struct vhost_virtqueue {/* Log write descriptors */ void __user *log_base;struct vhost_log *log; +/* delay multiple used buffers to signal once */ +int num_pend; +struct vring_used_elem *pend; };struct vhost_dev {
Quoted 4 lines · Shirley Ma
This patch changes vhost TX used buffer signal to guest from one by one to up to 3/4 of vring size. This change improves vhost TX message size from 256 to 8K performance for both bandwidth and CPU utilization without inducing any regression.
Any concerns about introducing latency or does the guest not care when TX completions come in?
Quoted 18 lines
Signed-off-by Shirley Ma <[email protected]>---
drivers/vhost/net.c | 19 ++++++++++++++++++-
drivers/vhost/vhost.c| 31 +++++++++++++++++++++++++++++++drivers/vhost/vhost.h| 3 +++3 files changed, 52 insertions(+), 1 deletions(-)drivers/vhost/net.c +0
-0@@ -198,7 +198,24 @@ static void handle_tx(struct vhost_net *net) if (err != len)pr_debug("Truncated TX packet: " " len %d != %zd\n", err, len);
-vhost_add_used_and_signal(&net->dev, vq, head, 0);
+ /*
Quoted 8 lines
+ * if no pending buffer size allocate, signal used buffer
+ * one by one, otherwise, signal used buffer when reaching
+ * 3/4 ring size to reduce CPU utilization.
+ */
+ if (unlikely(vq->pend))
+vhost_add_used_and_signal(&net->dev, vq, head, 0);
+ else {
+ vq->pend[vq->num_pend].id = head;
I don't understand the logic here: if !vq->pend then we assign to vq->pend[vq->num_pend].
Quoted 34 lines
+ vq->pend[vq->num_pend].len = 0;
+ ++vq->num_pend;
+ if (vq->num_pend== (vq->num- (vq->num>> 2))) {
+ vhost_add_used_and_signal_n(&net->dev, vq,
+vq->pend,
+vq->num_pend);
+vq->num_pend= 0;
+ }
+ }
total_len += len;
if (unlikely(total_len >= VHOST_NET_WEIGHT)) {vhost_poll_queue(&vq->poll);drivers/vhost/vhost.c +10
-0@@ -170,6 +170,16 @@ static void vhost_vq_reset(struct vhost_dev *dev, vq->call_ctx = NULL; vq->call = NULL; vq->log_ctx = NULL; + /* signal pending used buffers */ + if (vq->pend) { + if (vq->num_pend != 0) { + vhost_add_used_and_signal_n(dev, vq, vq->pend, + vq->num_pend); + vq->num_pend = 0; + } + kfree(vq->pend); + } + vq->pend = NULL; } static int vhost_worker(void *data)
drivers/vhost/vhost.c long vhost_dev_init(struct vhost_dev *dev, · line 292
Quoted 10 lines
drivers/vhost/vhost.c +6
-0@@ -273,7 +283,13 @@ long vhost_dev_init(struct vhost_dev *dev, dev->vqs[i].heads = NULL; dev->vqs[i].dev = dev; mutex_init(&dev->vqs[i].mutex); + dev->vqs[i].num_pend = 0; + dev->vqs[i].pend = NULL; vhost_vq_reset(dev, dev->vqs + i); + /* signal 3/4 of ring size used buffers */ + dev->vqs[i].pend = kmalloc((dev->vqs[i].num - + (dev->vqs[i].num >> 2)) * + sizeof *vq->peed, GFP_KERNEL);
Has this patch been compile tested? vq->peed?
Stefan
On Thu, Oct 28, 2010 at 9:57 AM, Stefan Hajnoczi <[email protected]> wrote:
Just read the patch 1/1 discussion and it looks like you're already on it. Sorry for the noise.
Stefan