Insufficient hugepages in the VM - thanks for everyone's help!
Ray K
On 19/02/2018 15:38, Ray Kinsella wrote:
Part of the mystery is solved in any case,
Looks to be related to running inside the VM.
DBGvpp# test tcp clients nclients 4 mbytes 256 test-timeout 100 uri
tcp://192.168.1.1/9000
4 three-way handshakes in 4.10 seconds .98/s
Test started at 10.538212
Test finished at 26.065026
1073741824 bytes (1024 mbytes, 1 gbytes) in 15.53 seconds
69154035.62 bytes/second full-duplex
.5532 gbit/second full-duplex
Ray K
On 14/02/2018 15:51, Florin Coras wrote:
Hi Ray,
The only thing missing with memif is TCO but that shouldn’t be a
reason for such a drop. I noticed you’re running a debug image, could
you try with release as well?
Cheers,
Florin
On Feb 14, 2018, at 7:42 AM, Ray Kinsella <m...@ashroe.eu
<mailto:m...@ashroe.eu>> wrote:
Hi Florin,
So I connected the two containers directly as Memif Master/Slave,
taking in VPP vSwitch completely out. Performance is double - but
still is pretty awful.
Could this be because I am not using DPDK under the hood in either
Container?
Ray K
DBGvpp# test tcp clients nclients 1 mbytes 16 test-timeout 100
uritcp://192.168.1.1/9000
1 three-way handshakes in .02 seconds 40.67/s
Test started at 308.999241
Test finished at 318.521999
16777216 bytes (16 mbytes, 0 gbytes) in 9.52 seconds
1761802.18 bytes/second full-duplex
.0141 gbit/second full-duplex
--------------- cone ---------------
DBGvpp# show error
Count Node Reason
23498 session-queue Packets transmitted
4 tcp4-rcv-process Packets pushed into
rx fifo
23498 tcp4-established Packets pushed into
rx fifo
4 ip4-icmp-input echo replies sent
1 arp-input ARP replies sent
DBGvpp# show ha
Name Idx Link Hardware
local0 0 down local0
local
memif0/0 1 up memif0/0
Ethernet address 02:fe:70:35:68:de
MEMIF interface
instance 0
--------------- ctwo ---------------
DBGvpp# show error
Count Node Reason
23522 session-queue Packets transmitted
2 tcp4-rcv-process Packets pushed into
rx fifo
23522 tcp4-established Packets pushed into
rx fifo
1 ip4-glean ARP requests sent
4 ip4-icmp-input unknown type
1 arp-input ARP request IP4
source address learned
DBGvpp# show ha
Name Idx Link Hardware
local0 0 down local0
local
memif0/0 1 up memif0/0
Ethernet address 02:fe:a3:b6:94:cd
MEMIF interface
instance 0
On 13/02/2018 16:37, Florin Coras wrote:
It would really help if read the whole email!
Apparently the test finishes, albeit with miserable performance! So,
for some reason lots and lots of packets are lost and that’s what
triggers the “heuristic” in the test client that complains the
connection is stuck. What does “show error” say? Does memif output
something for “show ha”?
Florin
On Feb 13, 2018, at 8:20 AM, Ray Kinsella <m...@ashroe.eu
<mailto:m...@ashroe.eu>> wrote:
Still stuck ...
DBGvpp# test tcp clients nclients 1 mbytes 16 test-timeout 100 uri
tcp://192.168.1.1/9000
1 three-way handshakes in .05 seconds 21.00/s
Test started at 205.983776
0: builtin_client_node_fn:216: stuck clients
Test finished at 229.687355
16777216 bytes (16 mbytes, 0 gbytes) in 23.70 seconds
707792.53 bytes/second full-duplex
.0057 gbit/second full-duplex
As a complete aside - pings appear to be slow enough through the
VPP vSwitch?
DBGvpp# ping 192.168.1.2
64 bytes from 192.168.1.2: icmp_seq=2 ttl=64 time=64.8519 ms
64 bytes from 192.168.1.2: icmp_seq=3 ttl=64 time=69.1016 ms
64 bytes from 192.168.1.2: icmp_seq=4 ttl=64 time=64.1253 ms
64 bytes from 192.168.1.2: icmp_seq=5 ttl=64 time=62.5618 ms
Ray K
On 13/02/2018 16:10, Florin Coras wrote:
test-timeout 100
-=-=-=-=-=-=-=-=-=-=-=-
Links:
You receive all messages sent to this group.
View/Reply Online (#8266): https://lists.fd.io/g/vpp-dev/message/8266
View All Messages In Topic (9): https://lists.fd.io/g/vpp-dev/topic/11145094
Mute This Topic: https://lists.fd.io/mt/11145094/21656
New Topic: https://lists.fd.io/g/vpp-dev/post
Change Your Subscription: https://lists.fd.io/g/vpp-dev/editsub/21656
Group Home: https://lists.fd.io/g/vpp-dev
Contact Group Owner: vpp-dev+ow...@lists.fd.io
Terms of Service: https://lists.fd.io/static/tos
Unsubscribe: https://lists.fd.io/g/vpp-dev/unsub
-=-=-=-=-=-=-=-=-=-=-=-
