The netperf  test result:

Test w/ virtio-net-pci:
# netperf -H 10.66.7.126 -t UDP_STREAM
MIGRATED UDP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 10.66.7.126 () port 0 AF_INET
Socket  Message  Elapsed      Messages                
Size    Size     Time         Okay Errors   Throughput
bytes   bytes    secs            #      #   10^6bits/sec

212992   65507   10.00      553431      0    29002.37
212992           10.00       19213           1006.85

Test w/ e1000:
# netperf -H 10.66.7.126 -t UDP_STREAM
MIGRATED UDP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 10.66.7.126 () port 0 AF_INET
Socket  Message  Elapsed      Messages                
Size    Size     Time         Okay Errors   Throughput
bytes   bytes    secs            #      #   10^6bits/sec

212992   65507   10.00       16211      0     849.53
212992           10.00        3863            202.44

When test the same netperf between guest and hostA, not hit this issue

Hi, Michael

I test this case w/ two hosts, that in the same broadcast domain, not hit this issue, both network of hosts works well during UDP_STREAM tests:

HostA:
# ethtool em1
Settings for em1:
	Supported ports: [ TP ]
	Supported link modes:   10baseT/Half 10baseT/Full 
	                        100baseT/Half 100baseT/Full 
	                        1000baseT/Full 
	Supported pause frame use: No
	Supports auto-negotiation: Yes
	Advertised link modes:  10baseT/Half 10baseT/Full 
	                        100baseT/Half 100baseT/Full 
	                        1000baseT/Full 
	Advertised pause frame use: No
	Advertised auto-negotiation: Yes
	Speed: 1000Mb/s
	Duplex: Full
	Port: Twisted Pair
	PHYAD: 2
	Transceiver: internal
	Auto-negotiation: on
	MDI-X: on
	Supports Wake-on: pumbg
	Wake-on: g
	Current message level: 0x00000007 (7)
			       drv probe link
	Link detected: yes
# ethtool -i em1
driver: e1000e
version: 2.2.14-k
firmware-version: 0.13-4
bus-info: 0000:00:19.0
supports-statistics: yes
supports-test: yes
supports-eeprom-access: yes
supports-register-dump: yes
supports-priv-flags: no


HostB:
# ethtool em1
Settings for em1:
	Supported ports: [ TP ]
	Supported link modes:   10baseT/Half 10baseT/Full 
	                        100baseT/Half 100baseT/Full 
	                        1000baseT/Full 
	Supported pause frame use: No
	Supports auto-negotiation: Yes
	Advertised link modes:  10baseT/Half 10baseT/Full 
	                        100baseT/Half 100baseT/Full 
	                        1000baseT/Full 
	Advertised pause frame use: No
	Advertised auto-negotiation: Yes
	Speed: 1000Mb/s
	Duplex: Full
	Port: Twisted Pair
	PHYAD: 2
	Transceiver: internal
	Auto-negotiation: on
	MDI-X: on
	Supports Wake-on: pumbg
	Wake-on: g
	Current message level: 0x00000007 (7)
			       drv probe link
	Link detected: yes

# ethtool -i em1
driver: e1000e
version: 2.2.14-k
firmware-version: 0.13-4
bus-info: 0000:00:19.0
supports-statistics: yes
supports-test: yes
supports-eeprom-access: yes
supports-register-dump: yes
supports-priv-flags: no


And the test result of netperf:
# netperf -H 10.66.5.221 -t UDP_STREAM
MIGRATED UDP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 10.66.5.221 () port 0 AF_INET
Socket  Message  Elapsed      Messages                
Size    Size     Time         Okay Errors   Throughput
bytes   bytes    secs            #      #   10^6bits/sec

212992   65507   10.00       18213      0     954.45
212992           10.00       18213            954.45


From the results, if this bug is a virt bug? 
If it's not a virt bug, pls feel free to change the components.

Best Regards!
Qian Guo

Just checking: this is not a regression right?
Same issue observed with 6.4?

also I am guessing this is with multiqueue disabled?

Please answer Michael's question and also try:

- Could it happens when you are doing pktgen from guest
- Does the problem disappear when adding sndbuf=1048576 in cli

Thanks

(In reply to jason wang from comment #6)
> Please answer Michael's question and also try:
> 
> - Could it happens when you are doing pktgen from guest
> - Does the problem disappear when adding sndbuf=1048576 in cli
> 
> Thanks
Since there's some urgent things in my hand, I will test assp after that, and update here.

Thanks!

(In reply to Michael S. Tsirkin from comment #4)
> Just checking: this is not a regression right?
> Same issue observed with 6.4?
Hi, Michael

I test w/ RHEL6.4GA, same issue, GuestA in HostA, and HostB run netserver, then during test UDP_STREAM, could not ping HostA from anywhere.

guest and host kernel:
# uname -r
2.6.32-358.el6.x86_64

qemu-kvm version:
# rpm -q qemu-kvm
qemu-kvm-0.12.1.2-2.355.el6.x86_64


And I did not test w/ multi-queue, so shall I test it in rhel7 w/ multi-queue?

I will test by comment #6, and update here.

Thanks !
Qian Guo

(In reply to jason wang from comment #6)
> Please answer Michael's question and also try:
> 
> - Could it happens when you are doing pktgen from guest
Test pktgen inside guest, i just use single queue
# ./pktgen.sh eth0 1
cat pktgen.sh

#!/bin/sh
# usage sh pktgen.sh $device $queues

modprobe -r pktgen
modprobe  pktgen
echo reset > /proc/net/pktgen/pgctrl

ifconfig $1 up

function pgset() {
    local result

    echo $1 > $PGDEV

    result=`cat $PGDEV | fgrep "Result: OK:"`
    if [ "$result" = "" ]; then
         cat $PGDEV | fgrep Result:
    fi
}

function pg() {
    echo inject > $PGDEV
    cat $PGDEV
}

for i in 0 `seq $(($2-1))`
do
    echo "Adding queue 0 of $1"
    dev=$1@$i

    PGDEV=/proc/net/pktgen/kpktgend_$i
    pgset "rem_device_all"
    pgset "add_device $dev"
    pgset "max_before_softirq 100000"

    # Configure the individual devices
    echo "Configuring devices $dev"

    PGDEV=/proc/net/pktgen/$dev

    pgset "queue_map_min $i"
    pgset "queue_map_max $i"
      pgset "count 10000000"
    pgset "min_pkt_size 60"
    pgset "max_pkt_size 60"
    pgset "dst 10.66.10.90"
    pgset "dst_mac 9a:05:06:07:08:10"
    pgset "udp_src_min 0"
    pgset "udp_src_max 65535"
    pgset "udp_dst_min 0"
    pgset "udp_dst_max 65535"
done

# Time to run

PGDEV=/proc/net/pktgen/pgctrl

echo "Running... ctrl^C to stop"

pgset "start"

echo "Done"


It works well, during pktgen test in guest, pinging this host well, no pkg lost.

> - Does the problem disappear when adding sndbuf=1048576 in cli
And no pkg lost when set sndbuf=1048576 and test UDP_STREAM netperf test.

It is the result when test w/ sndbuf=104876:

# netperf -H 10.66.6.254 -t UDP_STREAM -l 120
MIGRATED UDP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 10.66.6.254 () port 0 AF_INET
Socket  Message  Elapsed      Messages                
Size    Size     Time         Okay Errors   Throughput
bytes   bytes    secs            #      #   10^6bits/sec

212992   65507   120.09     218778      0     954.71
229376           120.09     218778            954.71


> 
> Thanks