190, 000 msgs/sec with very short avg response time (about 0.5 ms) for persistent connection case when I punch with only 100 concurrent client threads.
Mike McGrady
mmcgrady at topiatechnology.com
Sun Sep 13 11:25:56 EDT 2009
If you could provide the code, we could use it as a benchmark for
different machines. That is a suggestion. It would be handy to have
a simple benchmark application. This might be a start.
MIke
On Sep 13, 2009, at 8:18 AM, huican ping wrote:
> The input is only 55bytes for 190,000 TPS with 100% CPU usage..
> For 1024bytes input, the TPS I got is 88,500 msgs/sec with 89% CPU
> usage. (I cannot drives up the CPU usage anymore on server machine).
>
> The code is dummy, and I was using a simple decoder inherited from
> FrameDecoder, and really nothing special. It is for a tcp protocol,
> and there is just content, and no headers. I also use default worker
> thread number.
>
> I don't know what your protocol is for 76K, if you need to parse
> headers etc, it is pretty normal.
>
> Thanks
>
> On Sun, Sep 13, 2009 at 3:57 AM, Utkarsh Srivastava
> <utkarsh at gmail.com> wrote:
>> Could you post the code? Also, what was the size of each of your
>> messages?
>> On a dual core box, at 1K messages, I am only able to do 76K
>> messages per
>> sec.
>> Thanks
>> Utkarsh
>>
>> On Sat, Sep 12, 2009 at 9:27 PM, Huican Ping <pinghuican at gmail.com>
>> wrote:
>>>
>>> Hello netty users,
>>>
>>> I punched a simple netty server which decodes (FrameDecoder,
>>> checks its
>>> first and tailer bytes of the input) and returns message back with
>>> 55
>>> bytes
>>> input on my 2way Woodcrest machine: 2 core-2-dual at Intel(R)
>>> Xeon(R) CPU
>>> 5160 @3.00GHz
>>>
>>> The netty server can handle more than 190,000 msgs/sec with very
>>> short
>>> average response time (about 0.5 ms) for persistent connection
>>> case when I
>>> punch the server with 100 concurrent client threads.
>>>
>>> I got about 180,000 msgs/sec with about 0.28 ms avg response for
>>> persistent
>>> connection with 50 concurrent client threads.
>>>
>>> The client and server are different machines. They are in the same
>>> sub-network, but not directly connected on the switch. So the
>>> average
>>> response time can be really lower than above.
>>>
>>> client output:
>>> ==============
>>> |active| message | response time (ms) |avg conn|
>>> succ
>>> rt|
>>> connection |
>>> time(ms)|thrds | sent| succ| fail| min | avg | max
>>> |time(ms)|(msg/s)|pend|estb|idle|
>>> 10007 100 1863171 1863071 0 0.13 0.49 212.78
>>> 0.00
>>> 186176.78 0 100 0
>>> 20011 100 1901832 1901832 0 0.12 0.52 28.14
>>> 0.00
>>> 190126.16 0 100 0
>>> 30014 100 1904729 1904729 0 0.13 0.52 27.09
>>> 0.00
>>> 190434.81 0 100 0
>>> 40017 100 1901270 1901270 0 0.13 0.52 27.25
>>> 0.00
>>> 190069.98 0 100 0
>>> 50019 100 1906704 1906704 0 0.13 0.52 26.16
>>> 0.00
>>> 190632.27 0 100 0
>>> 60023 100 1906676 1906676 0 0.13 0.52 21.63
>>> 0.00
>>> 190610.42 0 100 0
>>>
>>> It is really impressive, decent and amazing number. A big thanks
>>> goes to
>>> Trustin Lee for his amazing work on Netty
>>>
>>> Thanks
>>> Huican
>>> --
>>> View this message in context:
>>> http://n2.nabble.com/190-000-msgs-sec-with-very-short-avg-response-time-about-0-5-ms-for-persistent-connection-case-when--tp3635215p3635215.html
>>> Sent from the Netty User Group mailing list archive at Nabble.com.
>>> _______________________________________________
>>> netty-users mailing list
>>> netty-users at lists.jboss.org
>>> https://lists.jboss.org/mailman/listinfo/netty-users
>>
>>
>> _______________________________________________
>> netty-users mailing list
>> netty-users at lists.jboss.org
>> https://lists.jboss.org/mailman/listinfo/netty-users
>>
>>
>
> _______________________________________________
> netty-users mailing list
> netty-users at lists.jboss.org
> https://lists.jboss.org/mailman/listinfo/netty-users
Mike McGrady
Principal Investigator AF081-028 AFRL SBIR
Senior Engineer
Topia Technology, Inc.
1.253.720.3365
mmcgrady at topiatechnology.com
More information about the netty-users
mailing list