VSP G200/G400/G600远程复制架构与概念白皮书（性能评估与灾备指南）

5星 · 超过95%的资源需积分: 13 141 浏览量更新于2024-07-15 收藏 2.63MB PDF 举报

本文档是《VSP G200、G400、G600远程复制架构与概念指南》（VSP G200 G400 G600 Remote Replication Architecture and Concepts Guide_v1.0.pdf），由Pamela Gray编写，出自Complex Test Lab - Open、Solutions Engineering和Technical Operations团队，专为Hitachi Data Systems合作伙伴及内部使用设计，非公开且需要签署NDA协议，发布日期为2015年11月。这份白皮书主要探讨了Hitachi Data Systems的高端存储系统系列VSP G200、G400和G600在远程数据复制（Remote Replication）方面的架构理念和技术概念。文档的核心内容涵盖了以下几个方面： 1. **远程复制架构**：详细介绍了VSP G200、G400和G600存储设备如何通过高效的数据传输技术实现远程站点之间的数据同步。这包括复制技术的原理、数据一致性保证机制、以及如何利用网络进行实时或周期性地备份数据，以确保业务连续性和灾难恢复能力。 2. **性能评估**：提供了存储系统的性能指标和基准测试结果，这些数据是在特定控制环境下获得的。然而，由于实际运行环境可能存在差异，读者需理解在其他环境中可能看到的性能表现可能会有所不同。这提示了用户在应用时需要根据自身环境进行调整和预期管理。 3. **注意事项和免责声明**：文档强调了性能数据的局限性，指出它是在孤立环境中获取的，不构成对其他环境的性能保证。同时，Hitachi Data Systems和供应商明确声明，产品和服务可能存在的所有设计、规格、声明、信息和建议（统称为“设计”）仅作为原始状态提供，不附带任何保证，包括但不限于适销性、特定用途的适用性、无侵犯知识产权的保证。 4. **最佳实践和建议**：白皮书中还包含了关于如何优化远程复制设置、减少数据传输延迟、提升数据安全性以及处理故障恢复策略的实用建议，帮助用户充分利用这些产品的功能来构建可靠和高效的容灾解决方案。这份指南提供了关于Hitachi VSP G200、G400和G600系列存储系统在远程复制技术上的深入见解，对于存储管理员和IT专业人员理解和实施远程数据保护策略具有重要的参考价值。但使用者在实际操作中应考虑环境差异，并结合具体业务需求制定相应的实施方案。

Hitachi Data Systems Partner and Internal Use Only – NDA Required Page 12

• 100% Cache Hit

o 100% Random Read

o 100% Random Write

o 100% Sequential Read

o 100% Sequential Write

• 100% Cache Miss

o 100% Random Read

o 100% Random Write

• Simulated OLTP Workload

o 20% cache hit, 75% 8 KB Random, 25% 256KB Sequential, 80% Read, 20% Write

For the 300 KM tests in the Paired state the following workloads were run:

• 100% Random Write Hit

• Simulated OLTP Workload

o 20% cache hit, 75% 8 KB Random, 25% 256KB Sequential, 80% Read, 20% Write

The test started at a low I/O rate depending on the workload - e.g. 20000 for the front-end write

workload. The I/O rate was incremented by a constant I/O rate until the ‘knee of the curve’ was

reached, or the final data points were ‘clustered together’. The ‘knee of the curve’ is the I/O rate at

which the response time increases dramatically or for random workloads, or the throughput stops

increasing for sequential workloads. When the final data points are ‘clustered together’, then the

hardware resources were saturated. The results of these tests would allow for the comparison of fixed

I/O rates to response time and to throughput.

The Vdbench workloads were run against the primary volumes in the following configurations:

• Simplex

• Paired at 0 KM

• Paired at 300 KM

The MCU and RCU were directly connected for the tests using FC links, and connected through and Anue distance simulator set

to 0 KM for the tests using iSCSI links.

Hitachi Data Systems Partner and Internal Use Only – NDA Required Page 13

Test results and observations

In this section we will describe what we learned from the testing. That is, where possible, we will

describe general observations that will apply across a variety of workloads, and in various situations. All

of our tests used HDP (Hitachi Dynamic Provisioning) pools backed by RAID6 LDEVs. Appendices A, B

and C describe the SAN configurations used for these tests.

Read workloads were not affected by TrueCopy

The use of TrueCopy had no effect on read workloads. As can be seen in the figures below host

response times were not affected by TrueCopy which means that the overhead of TrueCopy is negligible

for read operations as we would expect. Figures 1-3 show the results of a 100% random read hit

workload test at 0 KM. Because the 100% random read hit workload has no backend bottlenecks, it is

the workload that will show any TrueCopy overhead. As can be seen the response time of the SMPL

state and the PAIR state is identical at all I/O rates.

As can be seen from Figure 2 the use of iSCSI links does not change these results – TrueCopy still does

not affect read workloads. Also, the maximum I/O rate for FC and iSCSI links is the same showing that

TrueCopy is not a factor for read workloads. The limitation here is the MPU utilization.

The VSP G200 reaches a lower maximum I/O rate than the VSP G600 as we’d expect since the VSP G600

has 2 MPUs (4 virtual MPs) while the VSP G600 has 4 MPUs. The VSP G600 has a larger cache also. The

maximum I/O rate for the VSP G600 with iSCSI links and the VSP G200 were limited by the MPU

utilization, but the maximum I/O rate of the VSP G600 was limited by the hosts used to drive the I/O.

Therefore, with more powerful hosts the VSP G600 could have achieved an even higher maximum I/O

rate.

Figure 1. 100% random read hits at 0 KM for the VSP G600 using FC links

0 80000 160000 240000 320000 400000 480000 560000

IO Response Time (ms)

IOPS

VSP G600 TC with HDP - Open Systems Performance

Four HDP Pools, 72 RGs, 512 DP-Vol, Simplex vs. Pair

8K BLK, 100% Random, 100% Read, 100% Hit

Simplex Pair

70% MPU

Host

Saturation

70% MPU

剩余69页未读，继续阅读

zp989

粉丝: 0
资源: 1

VSP G200/G400/G600远程复制架构与概念白皮书（性能评估与灾备指南）

VSP_更换Cache电池

VSP_G400_G600_v83_05-2x_Hardware_Reference_MK-94HM8022-12.pdf

GX8010_VSP_SDK_DG.pdf

如何将def FC这个函数中所有averagespeed_vsp改为speed_vsp

VSP上下行波分离代码

利用fk滤波分离VSP上行波和下行波

HDSG600 存储告警消除

VSP处理地震数据的程序代码

gvim左右栏切换快捷键

VSP上下行波怎么分离，有没有具体的代码可以参考，分离效果比较好的

最新资源