failed to run ofp with nginx

[HsuJv]

Hi there,

I'm trying to run nginx with ofp to have a basic performance test wrk -c4000 -d120s -t12 --latency "http://myserver//"

my playing is:

vm os: centos 7.7.1908 (kernel 3.10.0-1062)
dpdk: dpdk-18.11.5
odp: odp 1.23.3
ofp: clone from this repository with the latest version
nginx: 1.16.1

I find that there is a feature netwrap to do this with the original Nginx #225 , but failed to run it.

./scripts/ofp_netwrap.sh ../nginx-1.16.1/objs/nginx -p ../nginx-1.16.1

....

D 3437958 3:3942635264 ofp_pkt_processing.c:174] ETH TYPE = 0800                                                                       
D 3437965 2:3951027968 ofp_pkt_processing.c:174] ETH TYPE = 0800                                                                       
D 3437971 3:3942635264 ofp_pkt_processing.c:174] ETH TYPE = 0069                                                                       
D 3437992 2:3951027968 ofp_pkt_processing.c:174] ETH TYPE = 86dd                                                                       
D 3437992 2:3951027968 ofp_pkt_processing.c:174] ETH TYPE = 86dd                                                                       
I 3438017 0:4160650240 app_main.c:216] End Netwrap processing constructor()                                                            

EAL: RTE_ACL tailq is already registered                                                                                               
PANIC in tailqinitfn_rte_acl_tailq():                                                                                                  
Cannot initialize tailq: RTE_ACL                                                                                                       
5: [/lib64/ld-linux-x86-64.so.2(+0x115a) [0x7ffff7ddc15a]]                                                                             
4: [/lib64/ld-linux-x86-64.so.2(+0xf973) [0x7ffff7dea973]]                                                                             
3: [/home/sonicwall/odp_ofp/lib/libofp.so.0.0.0(+0xd3b9c) [0x7ffff61dab9c]]                                                            
2: [/home/sonicwall/odp_ofp/lib/libofp_netwrap_crt.so.0.0.0(__rte_panic+0xba) [0x7ffff7066758]]                                        
1: [/home/sonicwall/odp_ofp/lib/libofp_netwrap_crt.so.0.0.0(rte_dump_stack+0x1a) [0x7ffff70dc2da]]                                     
./scripts/ofp_netwrap.sh: line 9: 35020 Aborted                 LD_LIBRARY_PATH=/path/tomy/odp_ofp/lib:$LD_LIBRARY_PATH LD_PRELOAD
=libofp_netwrap_crt.so.0.0.0:libofp.so.0.0.0:libofp_netwrap_proc.so.0.0.0 $@    

I also tried to set the Nginx to daemon off; master_process off;, failed either with the same stack.

Then I'm going to another project nginx_ofp. Again failed to run it.

[root@localhost nginx_dpdk]# ./start_nginx.sh 0                                                                                        |
Starting nginx on interface 0
Found 1 worker_processes
[root@localhost nginx_dpdk]# signal is :(null)
len : 1e, file : /usr/local/nginx_dpdk/ofp.conf

ODP system info
---------------
ODP API version: 1.23.3
CPU model:       
CPU freq (hz):   0
Cache line size: 0
Core count:      0

Running ODP appl: "webserver"
-----------------
IF-count:        1
Using IFs:       0

EAL: Detected 4 lcore(s)
EAL: Detected 1 NUMA nodes
EAL: Multi-process socket /var/run/dpdk/rte/mp_socket
EAL: No free hugepages reported in hugepages-1048576kB
EAL: Probing VFIO support...
EAL: PCI device 0000:02:01.0 on NUMA socket -1
EAL:   Invalid NUMA socket, default to 0
EAL:   probe driver: 8086:100f net_e1000_em
EAL: PCI device 0000:02:05.0 on NUMA socket -1
EAL:   Invalid NUMA socket, default to 0
EAL:   probe driver: 8086:100f net_e1000_em
Pool config:
  pool.pkt.max_num: 262143

Queue config:
  queue_basic.max_queue_size: 8191
  queue_basic.default_queue_size: 4095

Using scheduler 'basic'
Scheduler config:
  sched_basic.prio_spread: 4
  sched_basic.prio_spread_weight: 63
  sched_basic.burst_size_default[] =  32  32  32  32  32  16   8   4
  sched_basic.burst_size_max[] =     255 255 255 255 255  16  16   8
  sched_basic.group_enable.all: 1
  sched_basic.group_enable.worker: 1
  sched_basic.group_enable.control: 1

Packet IO config:
  pktio.pktin_frame_offset: 0

PKTIO: initialized null interface.
No crypto devices available
I 739911 0:4160634880 ofp_uma.c:45] Creating pool 'udp_inpcb', nitems=60000 size=1160 total=69600000
I 739914 0:4160634880 ofp_uma.c:45] Creating pool 'tcp_inpcb', nitems=60000 size=1160 total=69600000
I 739918 0:4160634880 ofp_uma.c:45] Creating pool 'tcpcb', nitems=60000 size=784 total=47040000
I 739921 0:4160634880 ofp_uma.c:45] Creating pool 'tcptw', nitems=60000 size=80 total=4800000
I 739923 0:4160634880 ofp_uma.c:45] Creating pool 'syncache', nitems=30720 size=168 total=5160960
I 739923 0:4160634880 ofp_uma.c:45] Creating pool 'tcpreass', nitems=320 size=48 total=15360
I 739923 0:4160634880 ofp_uma.c:45] Creating pool 'sackhole', nitems=65536 size=40 total=2621440
odp_crypto.c:598:odp_crypto_capability():No crypto devices available
I 739925 0:4160634880 ofp_init.c:438] Slow path threads on core 0
Num worker threads: 2
first CPU:          2
cpu mask:           0xC
D 739925 0:4160634880 ofp_ifnet.c:302] Interface '0' becomes 'fp0', port 0
DPDK interface (net_e1000_em): 0
  num_rx_desc: 128
  num_tx_desc: 256
  rx_drop_en: 0
D 739925 0:4160634880 ofp_ifnet.c:54] Interface '0' supports IPv4 TX checksum offload
D 739925 0:4160634880 ofp_ifnet.c:76] Interface '0' supports UDP TX checksum offload
D 739925 0:4160634880 ofp_ifnet.c:98] Interface '0' supports TCP TX checksum offload
E 739925 0:4160634880 ofp_ifnet.c:163] Failed to create output queues.
E 739925 0:4160634880 ngx_ofp_module.c:224] Failed to init interface 0

I set the nginx.conf to run only on a single core worker_processes 1;, but it seems the Nginx setup with CPU mask 0xc. That will not work since I'm having a network probe with E1000. Then I turned to Vmxnet3 but got this

D 12325 0:4160634880 ofp_ifnet.c:302] Interface '0' becomes 'fp0', port 0
odp_packet_dpdk.c:182:init_options():Invalid number of TX descriptors
odp_packet_dpdk.c:688:setup_pkt_dpdk():Initializing runtime options failed
../linux-generic/odp_packet_io.c:334:setup_pktio_entry():Unable to init any I/O type.
E 12325 0:4160634880 ofp_ifnet.c:23] odp_pktio_open failed
E 12325 0:4160634880 ngx_ofp_module.c:224] Failed to init interface 0

Also failed on the multi-queue.

Kindly advise how to run it, please.

P.S. I actually not matter how to run nginx with ofp on multi-cores. The thing I want is to make the configure worker_processes 1; works since my benchmarks are all on a single core.

P.S.2 Also, I succeeded to run the webserver with the help of #241 but it was too slow to compare with original nginx. So I have to figure out how to run nginx with ofp.

webserver:
wrk -c4000 -d120s -t12 --latency "http://myserver:2048//"
Running 2m test @ http://myserver:2048//
  12 threads and 4000 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency    59.02ms   78.34ms   1.99s    94.99%
    Req/Sec   489.13    224.24     1.90k    65.84%
  Latency Distribution
     50%   42.63ms
     75%   47.08ms
     90%   54.24ms
     99%  329.61ms
  467540 requests in 2.00m, 8.47MB read
  Socket errors: connect 2987, read 467540, write 0, timeout 242
Requests/sec:   3894.37
Transfer/sec:     72.26KB

original nginx:
wrk -c4000 -d120s -t12 --latency "http://myserver//"
Running 2m test @ http://myserver//
  12 threads and 4000 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency    43.40ms   79.56ms   2.00s    94.97%
    Req/Sec     2.96k     1.16k    5.97k    58.85%
  Latency Distribution
     50%   27.50ms
     75%   32.65ms
     90%   40.89ms
     99%  293.86ms
  2825754 requests in 2.00m, 452.60MB read
  Socket errors: connect 2987, read 0, write 0, timeout 49
Requests/sec:  23539.40
Transfer/sec:      3.77MB

[HsuJv]

Simple solved by change the call odp_cpu_count() to 1 in the src of nginx_ofp, close it.

--- a/src/event/modules/ngx_ofp_module.c                                   
+++ b/src/event/modules/ngx_ofp_module.c                                   
@@ -178,7 +178,7 @@ int my_webserver(int if_count, char **if_name)         
                exit(EXIT_FAILURE);                                        
        }                                                                  

-       core_count = odp_cpu_count();                                      
+       core_count = 1;                                                    
        num_workers = core_count;                                          

        if (params.core_count)                                             
@@ -211,7 +211,7 @@ int my_webserver(int if_count, char **if_name)         
         if (num_queues_str) pktin_param.num_queues = atoi(num_queues_str);

         odp_pktout_queue_param_init(&pktout_param);                       
-        pktout_param.num_queues = odp_cpu_count();                        
+        pktout_param.num_queues = 1;                                      
         pktout_param.op_mode = ODP_PKTIO_OP_MT_UNSAFE;                    

         for (i = 0; i < params.if_count; i++) {   

[HsuJv]

Reopen this. It seems that ths nginx with ofp has a very terrible performance.

I tried wrk from other host in the same subnet and got

--------------------------nginx_ofp--------------------------
[xxx@yyy ~]$ wrk -c4000 -d120s -t12 --latency "http://myserver//"
Running 2m test @ http://myserver//
  12 threads and 4000 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency    29.50ms   20.13ms 301.76ms   62.44%
    Req/Sec     1.51k   759.18     4.08k    64.01%
  Latency Distribution
     50%   19.28ms
     75%   53.42ms
     90%   58.77ms
     99%   68.88ms
  1439503 requests in 2.00m, 229.34MB read
  Socket errors: connect 2987, read 0, write 0, timeout 0
Requests/sec:  11988.07
Transfer/sec:      1.91MB

--------------------------original nginx--------------------------
[xxx@yyy ~]$ wrk -c4000 -d120s -t12 --latency "http://myserver//"
Running 2m test @ http://myserver//
  12 threads and 4000 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency    41.19ms  104.87ms   2.00s    96.46%
    Req/Sec     2.39k     1.25k    9.29k    68.53%
  Latency Distribution
     50%   24.46ms
     75%   29.89ms
     90%   38.71ms
     99%  494.92ms
  2559265 requests in 2.00m, 409.92MB read
  Socket errors: connect 2987, read 0, write 0, timeout 945
Requests/sec:  21313.51
Transfer/sec:      3.41MB

Then I tried wrk from the host which runs nginx itself and got

--------------------------nginx_ofp--------------------------
[root@localhost wrk-master]# ./wrk -c1000 -d120s -t12 --latency "http://myserver//"                                               
Running 2m test @ http://myserver//
  12 threads and 1000 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency    55.50ms   10.34ms 268.90ms   87.49%
    Req/Sec     1.34k   214.35     2.35k    69.42%
  Latency Distribution
     50%   55.35ms
     75%   59.35ms
     90%   63.86ms
     99%   81.24ms
  1926373 requests in 2.00m, 306.85MB read
  Socket errors: connect 0, read 1, write 0, timeout 0
Requests/sec:  16040.38
Transfer/sec:      2.56MB

--------------------------original nginx--------------------------
[root@localhost wrk-master]# ./wrk -c1000 -d120s -t12 --latency "http://myserver//"                                               
Running 2m test @ http://myserver//
  12 threads and 1000 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency    30.68ms  106.45ms   2.00s    91.86%
    Req/Sec     3.27k     1.58k   21.29k    77.98%
  Latency Distribution
     50%    2.73ms
     75%    3.01ms
     90%   82.72ms
     99%  380.81ms
  4664706 requests in 2.00m, 747.14MB read
  Socket errors: connect 0, read 826, write 0, timeout 376
Requests/sec:  38840.71
Transfer/sec:      6.22MB

Here's my nginx.conf, both of the original and the ofp one use the same configure

user  root;
worker_processes  1;

error_log  logs/error.log  info;

events {
    worker_connections  1024;
}

http {
    include       mime.types;
    default_type  application/octet-stream;

    sendfile        off;

    keepalive_timeout  65;

    server {
        listen       80;
        server_name  localhost;

        access_log /dev/null;

        location / {
            return 200 "Testing";
        }

        error_page   500 502 503 504  /50x.html;
        location = /50x.html {
            root   html;
        }

    }

}

Thanks in advance for your kindly advise

[bogdanPricope]

Hi HsuJv,

Maybe I can help you with the 'netwrap' thing...

Take a look at https://github.com/OpenFastPath/ofp/blob/master/example/README (the netwrap section). Now, the "Restrictions" section is outdated. You will need to modify the configuration command:

1. disable slowpath: --enable-sp=no
2. enable netwrap flavor: --with-config-flv=netwrap-webserver e.g. ./configure --with-odp=...... --enable-sp=no --with-config-flv=netwrap-webserver

Next, take a look at script itself (https://github.com/OpenFastPath/ofp/blob/master/scripts/ofp_netwrap.sh): it uses by default 'eth1' and 'ofp_netwrap.cli' configuration file (192.168.100.1). You may need to adapte it to your needs (your interface, etc)

[HsuJv]

Hi bogdanPricope,

Thanks for your help. I've exactly tried the things you mentioned before, following the guidelines of #225 but still the issue.

But the netwrap thing is not the thing I'm concerned now, I finally made it work with nginx_ofp successfully as said in my comment 2.

And now I'm concerned about the performance issue as said in my comment 3. That's why I reopen this issue after closing it once.

Regards.

[QinLongFei]

Hi, bogdanPricope ,

I also care about the performance test between nginx and ofp-nginx unber same hardware config.

Do we have official performance test about ofp-nginx?

Thanks

[bogdanPricope]

Sorry .. I cannot help you with ofp-nginx.

[HsuJv]

Never mind, I'll leave this issue open to wait someone help.

Anyway thanks for your kindly help.