servicecomb-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "新道场开张了" <tsubasa...@qq.com>
Subject Re: Saga transaction performance test with Sharding-JDBC
Date Tue, 25 Dec 2018 08:31:21 GMT
XA use Atomikos as default.


The test run with 
repo: https://github.com/cherrylzhao/sharding-sphere-example/tree/dev-performance
module: sharding-jdbc-example --> spring-boot-nodep-example --> spring-boot-nodep-mybatis-example
class: io.shardingsphere.example.spring.boot.mybatis.nodep.SpringBootStarterTransactionExample


------------------ Original ------------------
From:  "Zheng Feng"<zh.feng@gmail.com>;
Date:  Tue, Dec 25, 2018 04:05 PM
To:  "dev"<dev@servicecomb.apache.org>;

Subject:  Re: Saga transaction performance test with Sharding-JDBC



Thanks for sharing these performance tests and is it possible to share the
source codes if you are happy to open these tests ?
In term of the XA tests, I think the default transaction manager is
Atomikos ? I'm interested with running with the Narayana [1]

So it could be very useful to share the source codes with the community !
Thanks,

Zheng Feng

[1] https://github.com/zhfeng/narayana-sharding-sphere

新道场开张了 <tsubasaotl@qq.com> 于2018年12月25日周二 上午10:51写道:

> Saga Tx is slower than No Tx.
> TPS of saga is 1700 and No Tx is 2000 in test environment.
>
>
> About the load and memory usage, we will record in next performance.
>
>
>
>
> ------------------ Original ------------------
> From:  "Zhang Yonglun"<zhangyonglun@apache.org>;
> Date:  Mon, Dec 24, 2018 08:56 PM
> To:  "dev"<dev@shardingsphere.apache.org>;
> Cc:  "dev"<dev@servicecomb.apache.org>;
> Subject:  Re: Saga transaction performance test with Sharding-JDBC
>
>
>
> Impressive!
> I am focused on the performance issue of ShardingSphere for a long time,
> and have done a little saga work before. But I still can't understand why
> saga Tx faster than No Tx. Is there something I missed?
>
> BTW, I noticed that second nice machine, and wonder what's the load and
> memory usage on it when testing.
>
>
> 新道场开张了 <tsubasaotl@qq.com> 于2018年12月24日周一 下午6:28写道:
>
> > Hi, everyone.
> >
> >
> > The feature of saga transaction in ShardingSphere has been basically
> > completed
> > by integrating servicecomb-saga-actuator.
> > Recently, we used Sharding-JDBC to test the performance of Saga
> > transactions.
> >
> >
> > There are results for two kinds of environments.
> >
> >
> > First result comes from local environment which including 2 cores and 16G
> > RAM
> > The connection pool size, thread pool size of saga-actuator and thread
> > pool size of Sharding-JDBC all are 50.
> >
> >
> > ---------------- result for local environment ----------------
> > |Tx Type|Thread Number|Average response(ms)| TPS |
> > | No Tx |      50     |         337        | 140 |
> > | saga  |      50     |         395        | 120 |
> > | local  |      50      |         323         |143|
> > | xa     |      50      |         301         |154|
> > | No Tx |     100     |         605        | 158 |
> > | saga  |     100     |         789        | 120 |
> >
> > ---------------- result for local environment ----------------
> >
> >
> > Second result comes from test environment which 256cores and 300+G RAM
> > The connection pool size, thread pool size of saga-actuator and thread
> > pool size of Sharding-JDBC all are 200.
> >
> >
> > ---------------- result for test environment ----------------
> > |Tx Type|Thread Number|Average response(ms)| TPS |
> > | No Tx |     200     |         95         |2002|
> > | saga  |      200     |         351       |1700|
> >
> > | local  |      200     |         64         |2868|
> > | xa     |      200     |         98         |2012|
> > ---------------- result for test environment ----------------
> >
> >
> > And I do echo test with emptyTransport which do not execute SQL in saga,
> > the result is TPS 457 in local and 3200 in test environment.
> >
> >
> > if saga do persistence to log file,  the TPS of saga will nose dive to 70
> > in local and 600+ in test environment
>
>
>
> --
> Zhang Yonglun
> Apache ShardingSphere
Mime
  • Unnamed multipart/alternative (inline, 8-Bit, 0 bytes)
View raw message