[Oisf-users] Cant start suricata with cuda in daemon mode

stefan.egger at bit.admin.ch stefan.egger at bit.admin.ch
Thu Dec 5 15:12:56 UTC 2013


Issue #1059 created.  

Regards Stefan 



-----Urspr√ľngliche Nachricht-----
Von: Anoop Saldanha [mailto:anoopsaldanha at gmail.com] 
Gesendet: Donnerstag, 5. Dezember 2013 16:02
An: Egger Stefan BIT
Betreff: Re: [Oisf-users] Cant start suricata with cuda in daemon mode

Stefan,

Missed the daemon mode part.  Haven't tried cuda + daemon mode yet.
Can you please open a bug report on
https://redmine.openinfosecfoundation.org/projects/suricata ?

On Thu, Dec 5, 2013 at 8:27 PM,  <stefan.egger at bit.admin.ch> wrote:
> Hi Anoop (hope that is your firstname)
>
> Yes I can give you some extra information:
>
> - Running suricata 2.0 Beta 1
> - SDK 5.5
> - centos 6.4
>
> #build info:
> ****************************************************************************
> [root at gandalf log]# /opt/suricata2/bin/suricata --build-info
> This is Suricata version 2.0beta1 RELEASE
> Features: UNITTESTS PCAP_SET_BUFF LIBPCAP_VERSION_MAJOR=1 CUDA AF_PACKET HAVE_PACKET_FANOUT LIBCAP_NG LIBNET1.1 HAVE_HTP_URI_NORMALIZE_HOOK HAVE_NSS HAVE_LIBJANSSON PROFILING
> 64-bits, Little-endian architecture
> GCC version 4.4.7 20120313 (Red Hat 4.4.7-3), C version 199901
>   __GCC_HAVE_SYNC_COMPARE_AND_SWAP_1
>   __GCC_HAVE_SYNC_COMPARE_AND_SWAP_2
>   __GCC_HAVE_SYNC_COMPARE_AND_SWAP_4
>   __GCC_HAVE_SYNC_COMPARE_AND_SWAP_8
>   __GCC_HAVE_SYNC_COMPARE_AND_SWAP_16
> L1 cache line size (CLS)=64
> compiled with LibHTP v0.5.5, linked against LibHTP v0.5.5
> Suricata Configuration:
>   AF_PACKET support:                       yes
>   PF_RING support:                         no
>   NFQueue support:                         no
>   IPFW support:                            no
>   DAG enabled:                             no
>   Napatech enabled:                        no
>   Unix socket enabled:                     yes
>
>   libnss support:                          yes
>   libnspr support:                         yes
>   libjansson support:                      yes
>   Prelude support:                         no
>   PCRE jit:                                no
>   libluajit:                               no
>   libgeoip:                                yes
>   Non-bundled htp:                         no
>   Old barnyard2 support:                   no
>   CUDA enabled:                            yes
>
>   Suricatasc install:                      yes
>
>   Unit tests enabled:                      yes
>   Debug output enabled:                    no
>   Debug validation enabled:                no
>   Profiling enabled:                       yes
>   Profiling locks enabled:                 no
>
> Generic build parameters:
>   Installation prefix (--prefix):          /opt/suricata2
>   Configuration directory (--sysconfdir):  /opt/suricata2/etc/suricata/
>   Log directory (--localstatedir) :        /opt/suricata2/var/log/suricata/
>
>   Host:                                    x86_64-unknown-linux-gnu
>   GCC binary:                              gcc
>   GCC Protect enabled:                     no
>   GCC march native enabled:                yes
>   GCC Profile enabled:                     no
> [root at gandalf log]#
>
> cuda parameter from the yaml file:
> ****************************************************************************
> # Cuda configuration.
>
> cuda:
>   # The "mpm" profile.  On not specifying any of these parameters, the engine's
>   # internal default values are used, which are same as the ones specified in
>   # in the default conf file.
>   mpm:
>     # The minimum length required to buffer data to the gpu.
>     # Anything below this is MPM'ed on the CPU.
>     # Can be specified in kb, mb, gb.  Just a number indicates it's in bytes.
>     # A value of 0 indicates there's no limit.
>     data-buffer-size-min-limit: 40
>     # The maximum length for data that we would buffer to the gpu.
>     # Anything over this is MPM'ed on the CPU.
>     # Can be specified in kb, mb, gb.  Just a number indicates it's in bytes.
>     data-buffer-size-max-limit: 1500
>     # The ring buffer size used by the CudaBuffer API to buffer data.
>     cudabuffer-buffer-size: 500mb
>     # The max chunk size that can be sent to the gpu in a single go.
>     #ORIG: gpu-transfer-size: 50mb
>     gpu-transfer-size: 10mb
>     # The timeout limit for batching of packets in microseconds.
>     batching-timeout: 2000
>     # The device to use for the mpm.  Currently we don't support load balancing
>     # on multiple gpus.  In case you have multiple devices on your system, you
>     # can specify the device to use, using this conf.  By default we hold 0, to
>     # specify the first device cuda sees.  To find out device-id associated with
>     # the card(s) on the system run "suricata --list-cuda-cards".
>     device-id: 0
>     # No of Cuda streams used for asynchronous processing. All values > 0 are valid.
>     # For this option you need a device with Compute Capability > 1.0.
>     cuda-streams: 2
>
>
> Startup:
> ****************************************************************************
> 5/12/2013 -- 15:41:53 - <Info> - allocated 3670016 bytes of memory for the defrag hash... 65536 buckets of size 56
> 5/12/2013 -- 15:41:53 - <Info> - preallocated 65535 defrag trackers of size 144
> 5/12/2013 -- 15:41:53 - <Info> - defrag memory usage: 13107056 bytes, maximum: 33554432
> 5/12/2013 -- 15:41:53 - <Info> - AutoFP mode using default "Active Packets" flow load balancer
> 5/12/2013 -- 15:41:53 - <Info> - Use pid file /opt/suricata2/var/run/suricata.pid from config file.
> 5/12/2013 -- 15:41:54 - <Info> - preallocated 60000 packets. Total memory 1520640000
> 5/12/2013 -- 15:41:54 - <Info> - allocated 229376 bytes of memory for the host hash... 4096 buckets of size 56
> 5/12/2013 -- 15:41:54 - <Info> - preallocated 1000 hosts of size 120
> 5/12/2013 -- 15:41:54 - <Info> - host memory usage: 349376 bytes, maximum: 16777216
> 5/12/2013 -- 15:41:54 - <Info> - allocated 3670016 bytes of memory for the flow hash... 65536 buckets of size 56
> 5/12/2013 -- 15:41:54 - <Info> - preallocated 10000 flows of size 272
> 5/12/2013 -- 15:41:54 - <Info> - flow memory usage: 6390016 bytes, maximum: 33554432
> 5/12/2013 -- 15:41:54 - <Info> - IP reputation disabled
> 5/12/2013 -- 15:41:54 - <Info> - Added "34" classification types from the classification file
> 5/12/2013 -- 15:41:54 - <Info> - Added "12" reference types from the reference.config file
> 5/12/2013 -- 15:41:54 - <Info> - using magic-file /usr/share/file/magic
> 5/12/2013 -- 15:41:54 - <Info> - Delayed detect disabled
> 5/12/2013 -- 15:41:55 - <Info> - 38 rule files processed. 7822 rules successfully loaded, 0 rules failed
> 5/12/2013 -- 15:41:59 - <Info> - 7830 signatures processed. 567 are IP-only rules, 3444 are inspecting packet payload, 4942 inspect application layer, 72 are decoder event only
> 5/12/2013 -- 15:41:59 - <Info> - building signature grouping structure, stage 1: adding signatures to signature source addresses... complete
> 5/12/2013 -- 15:41:59 - <Info> - building signature grouping structure, stage 2: building source address list... complete
> 5/12/2013 -- 15:42:01 - <Info> - building signature grouping structure, stage 3: building destination address lists... complete
> 5/12/2013 -- 15:42:01 - <Error> - [ERRCODE: SC_ERR_CUDA_ERROR(134)] - cuCtxPushCurrent failed.  Returned errocode - CUDA_ERROR_NOT_INITIALIZED
> 5/12/2013 -- 15:42:01 - <Error> - [ERRCODE: SC_ERR_FATAL(171)] - context push failed.
>
>
> cardinfo:  Nvidia Quadro K4000 with 3GB RAM
> ****************************************************************************
>
> [root at gandalf log]# /opt/suricata2/bin/suricata --list-cuda-cards
> 5/12/2013 -- 15:48:41 - <Info> - This is Suricata version 2.0beta1 RELEASE
> 5/12/2013 -- 15:48:41 - <Info> - CPUs/cores online: 40
> 5/12/2013 -- 15:48:42 - <Info> - GPU Device 1: Quadro K4000, 4 Multiprocessors, 810MHz, CUDA Compute Capability 3.0
> CUDA Cards recognized by the suricata CUDA module -
> |-----------------------------------------------------------------------------|
> | Device Id  |     Device Name      |   Multi-   | Clock Rate | Cuda Compute  |
> |            |                      | Processors |    (MHz)   | Capability    |
> |-----------------------------------------------------------------------------|
> | 0          | Quadro K4000         | 4          | 810        | 3.0           |
> |-----------------------------------------------------------------------------|
> [root at gandalf log]#
>
>
> Change the Memory parameter did not solve the problem. Just keep in mind, that it's running well
> without daemon mode ...
>
> Regards Stefan
>
>
>
>
>
> -----Urspr√ľngliche Nachricht-----
> Von: Anoop Saldanha [mailto:anoopsaldanha at gmail.com]
> Gesendet: Donnerstag, 5. Dezember 2013 15:39
> An: Egger Stefan BIT
> Cc: oisf-users at lists.openinfosecfoundation.org
> Betreff: Re: [Oisf-users] Cant start suricata with cuda in daemon mode
>
> On Thu, Dec 5, 2013 at 6:51 PM,  <stefan.egger at bit.admin.ch> wrote:
>> Hi all,
>>
>> has anyone problems starting suricata with a cuda card in daemon mode?
>> For some reason I can't start suricata. I'm running into a cude error .
>> If I start normal I don't get an error everything seem to be fine.
>>
>> 5/12/2013 -- 14:17:10 - <Info> - building signature grouping structure, stage 1: adding signatures to signature source addresses...
>> complete
>> 5/12/2013 -- 14:17:10 - <Info> - building signature grouping structure, stage 2: building source address list... complete
>> 5/12/2013 -- 14:17:11 - <Info> - building signature grouping structure, stage 3: building destination address lists... complete
>> 5/12/2013 -- 14:17:11 - <Error> - [ERRCODE: SC_ERR_CUDA_ERROR(134)] - cuCtxPushCurrent failed.  Returned errocode -
>> CUDA_ERROR_NOT_INITIALIZED
>> 5/12/2013 -- 14:17:11 - <Error> - [ERRCODE: SC_ERR_FATAL(171)] - context push failed.
>>
>>
>> Any advice would be welcomed
>
> 1. Do you have any more startup logs?
> 2. Can you give me your card specs?
> 3. What is the version of cuda runtime you are using?
> 4. In the yaml file there are some cuda settings.  The below parameter -
>
>     # The max chunk size that can be sent to the gpu in a single go.
>     gpu-transfer-size: 50mb
>
> Can you reduce the size to a smaller value(10mb) and see if that
> solves the issue?
>
> --
> -------------------------------
> Anoop Saldanha
> http://www.poona.me
> -------------------------------



-- 
-------------------------------
Anoop Saldanha
http://www.poona.me
-------------------------------
-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/x-pkcs7-signature
Size: 8930 bytes
Desc: not available
URL: <http://lists.openinfosecfoundation.org/pipermail/oisf-users/attachments/20131205/1e94c523/attachment-0002.bin>


More information about the Oisf-users mailing list