[Question] 从1.3.0单节点模式升级到1.5.0集群模式性能并未提升

[liqigui]

Problem Type (问题类型)

gremlin (结果不合预期)

Before submit

• 我已经确认现有的 Issues 与 FAQ 中没有相同 / 重复问题 (I have confirmed and searched that there are no similar problems in the historical issue and documents)

Environment (环境信息)

机器配置如下：
内存: 32g
cpu核心数:16
cpu架构:aarch64

测试数据规模:
顶点 34w
边 35w

Your Question (问题描述)

1.5集群模式对比1.3.0单节点测试测试结果如下:
1.写入性能:比单节点慢50%
2.读性能:相同的语句(6到7层查询)，比1.3.0单节点慢10倍，1.3.0只需要600毫秒，1.5.0需要长达6秒

我们集群模式的安装方式是参考一下链接安装，目前都是部署在一台机器上面
https://github.com/apache/incubator-hugegraph/wiki/HugeGraph-Distributed-(pd%E2%80%90store)-Version-Deployment-Guide

机器配置如下：
内存: 32g
cpu核心数:16
cpu架构:aarch64

测试数据规模:
顶点 34w
边 35w

请问是不是哪里的配置不对呢？集群模式下性能会更差么？

Vertex/Edge example (问题点 / 边数据举例)

No response

Schema [VertexLabel, EdgeLabel, IndexLabel] (元数据结构)

No response

[dosubot[bot]]

@liqigui 你好！我是Dosu，一个可以帮助你解决 bug、回答问题并成为贡献者的机器人助手。我会在这里帮助你，等待人类维护者的到来。期待与你一起解决问题！🤖

I found a similar open issue titled "[Question] 从1.3.0单节点模式升级到1.5.0集群模式性能并未提升" which is closed and was last updated on December 23, 2024. However, this issue has no comments or references that provide a solution or further insights ^[1].

For improving performance in HugeGraph 1.5.0 cluster mode, consider the following configuration settings that could impact performance:

1. Cache Configuration:

   • Increase vertex.cache_capacity and edge.cache_capacity to improve read performance by reducing backend store access.
   • Adjust vertex.cache_expire and edge.cache_expire to manage cache expiration times.

2. Raft Configuration:

   • Ensure raft.use_replicator_pipeline is set to true to allow parallel log sending, which can enhance write performance.
   • Adjust raft.backend_threads and raft.read_index_threads based on your hardware capabilities to improve both read and write performance.

3. Task Configuration:

   • Optimize task.scheduler_type, task.schedule_period, and other task-related settings to better manage task execution and retries.

4. Backend Configuration:

   • Ensure the backend storage type (backend=hstore) and serializer (serializer=binary) are correctly configured for your environment.

These settings can be adjusted based on your specific workload and hardware capabilities to optimize performance ^[2][3][4][5]. Additionally, ensure that your backend configurations (e.g., RocksDB, Cassandra, HBase) are optimized for your environment, as these can significantly impact performance.

To continue talking to Dosu, mention @dosu.

---

^{Help Dosu learn! Give it feedback: Great Response | Irrelevant Answer | Incorrect Sources | Too Verbose | Hallucination | Other}

[JackyYangPassion]

首先两个方面进行

1. 写入性能
   a. 首先看下Server Store 内存配置多少
   b. 看下主机CPU 使用情况
   c. Store 是否在进行Flush /Compact 等 这个也会严重影响写入

2. 查询性能

读性能:相同的语句(6到7层查询)，比1.3.0单节点慢10倍，1.3.0只需要600毫秒，1.5.0需要长达6秒

采用 arthas 进行分析，看是慢在了哪里：计算层还是存储层

[JackyYangPassion]

#2611
这个监控相关的 DashBoard 可以参考，有助于分析定位问题

@liqigui