HPE 3PAR 8400 Configuration: Principles and Procedures

2026年8月26日 LOG admin

HPE 3PAR StoreServ 8400 是 8000 系列中高端存储阵列，12Gbps SAS 后端、16Gbps FC 前端、第五代 ASIC，最多 4 控制器集群。本文梳理 3PAR 的硬件组成（控制器框/扩展盘柜/Cage）、管理方式（SSMC 与 CLI）以及核心概念（CPG、虚拟卷 VV、VLUN），并给出创建虚拟卷、导出/取消导出、删除卷的完整流程，适合存储运维入门。

硬件架构

控制器框（Controller Enclosure / Base Storage System）：2U，前 24 盘位，后双控制器；
8400 支持扩展到 4 节点集群（两个控制器框）；单控制器框最多 11 个盘柜，双框 22 个；
每控制器：6 核 2.2GHz、16GB 缓存、2×12Gbps SAS、2×16Gbps FC、1×PCIe 扩展槽、2×1Gbps 管理/RCIP；
扩展卡可选 4×16G FC、2×10G iSCSI/FCoE、4×1G 或 2×10G 以太网。

Hardware Architecture in Practice

The 8000 series is built around a pair of independent controller nodes that are not a cluster in the traditional sense. Each node has its own CPU, cache and battery-backed write cache, and the two are connected by a high-speed internal PCIe fabric. Because both nodes can see every drive enclosure, either one can serve I/O for any volume — that is what makes controller failover transparent to hosts.

Three hardware concepts cover almost all daily work on an 8400:

Controller enclosure — the 2U base unit that holds the two controllers and the first 24 drive slots. Anything you add later is capacity only; there is no way to add a third controller to an existing pair without forming a new two-node pair and creating a four-node cluster.
Drive enclosure (cage) — an expansion shelf containing drives and a SAS loop. Each 8400 controller enclosure can attach up to eleven drive enclosures, and a four-node cluster doubles that. SAS loops are cabled as a pair, so a single cable failure does not remove the enclosure.
Cage — the internal grouping of drives within an enclosure, addressed by cage in CLI output. When you read disk health output, the cage number identifies which shelf and which position the drive sits in.

Front-end connectivity is defined by PCIe cards in each node, and the model matters: a 4×16G Fibre Channel card in a controller only accounts for half the ports you need, because each node carries its own cards. Always plan for ports from both controllers when sizing host connectivity, so a controller takeover does not strand a host path.

管理方式

SSMC（StoreServ Management Console）是现代 Web 管理界面，功能覆盖传统 MC 并支持性能监控与报表；CLI 通过 SSH 登录，默认用户 3paradm，常用命令：

hpe3par cli% shownode          # 控制器信息
hpe3par cli% checkhealth       # 阵列健康检查
hpe3par cli% showversion       # 固件版本
hpe3par cli% showcpg           # 查看 CPG

Both interfaces talk to the same underlying array, but they are not equivalent in everyday use. SSMC is better for capacity planning, performance trend charts and volume creation wizards; the CLI is faster for scripted checks, bulk operations and anything you want to put in a monitoring system. A third option, the REST API, exposes the same objects and is the right choice for automation.

cli% showversion -b             # both nodes and both controller cages
cli% shownode
cli% showport
cli% showhost
cli% showvv -d

For day-2 operations, three habits pay off. Log in as 3paradm only for administration and create a separate read-only account for monitoring. Capture checkhealth output on a schedule and store it, so you have a baseline to compare against. And keep firmware versions aligned between both controllers and the drive enclosure firmware, since mismatched levels are a common source of obscure performance problems.

核心概念

CPG（Common Provisioning Group）：按 RAID 级别与盘类型定义的容量池；
虚拟卷 VV：CPG 内的逻辑空间，可分 Full Provisioned（如日志盘）与 Thin Provisioned（如虚拟化），SSD 上可开 Thin Dedup；
VLUN：把 VV 导出给主机（映射 LUN）的机制。

Capacity Planning and RAID Choices

A Common Provisioning Group is the bridge between physical drives and logical volumes: it groups drives of one type (fast SSD, high-capacity NL, 10K SAS) under one RAID setting and presents a pool of space. Volumes are carved from a CPG and inherit its RAID level and its performance characteristics, which is why getting the CPG layout right at install time is worth more than any later tuning.

Practical guidance that holds for most 8400 deployments:

RAID 1 (or 10) for anything latency-sensitive — database logs, boot volumes, write-heavy OLTP. You pay 50% capacity for the best write behaviour.
RAID 5 for general-purpose virtual machine datastores. It offers the best capacity efficiency at a tolerable write penalty on spinning media.
RAID 6 for large, low-IOPS archives and backup targets where a double drive failure during rebuild must not cost you data.
Separate CPGs for SSD and NL even in an adaptive flash cache design, so you can move a volume between performance tiers without touching its WWN.
Thin provisioning is the default; use fully provisioned volumes only where the host or application genuinely needs a fixed LUN size and cannot tolerate a capacity alarm.

Watch the raw to usable ratio. With RAID 6 and a large drive count the parity overhead is modest, but on a small drive set it can be far more expensive than the numbers on the data sheet suggest. Model the layout in SSMC before committing to a CPG design.

虚拟卷操作流程

创建虚拟卷

SSMC：顶部 3PAR StoreServ → Block Persona → Virtual Volumes → Create virtual volume，向导内可同时完成 Export。CLI 也可用 createvv 系列命令。

cli% createvv -s 1024 CPG_SSD VV_WEB01
cli% createvv -s 512 CPG_NL -tpvv VV_ARCHIVE01
cli% showvv -d VV_WEB01

The -s parameter sets the size in MiB by default, -tpvv requests a thinly provisioned volume, and omitting it creates a fully provisioned one. Adding -tdvv enables thin deduplication, which is only available on SSD-backed CPGs and is worth switching on for volumes holding many similar virtual machines.

导出与取消导出

选择目标卷 → Exports 视图 → Actions → Unexport，确认删除 VLUN。

cli% createvlun VV_WEB01 1 HOST_WEB01
cli% showvlun -a
cli% removevlun VV_WEB01 1 HOST_WEB01

The LUN number is what the host sees, so changing it later requires a rescan and often a reconfiguration of the host's multipath software. Export the volume to a host set rather than to individual hosts whenever the same volume must reach a cluster — it keeps future node additions to a single command.

删除虚拟卷

先确认无主机使用，删除时系统会二次确认数据将被清除。

cli% showvlun -v VV_WEB01
cli% removevv VV_WEB01
cli% removevv -f VV_WEB01

Before removing a volume, check three things: no VLUN still maps it, no snapshot or Remote Copy group references it, and no host still has a filesystem mounted from it. The -f flag bypasses part of the safety check and should be treated as a last resort, not as a convenience.

Host Connectivity and Multipathing

Presenting a volume is only half the job; the host side decides whether the array's redundancy actually helps you. On an 8400 with two controllers and four front-end ports, each host should see at least two paths, one through each controller.

Fibre Channel — zone one host HBA port to one array port, repeat for each pair, and let multipath software (Native Multipathing, PowerPath/MPIO, or the 3PAR Host Explorer agent) aggregate them. Zone changes should be done one fabric at a time.
iSCSI — use the array's showport output to confirm which ports are enabled for iSCSI, then create the host object with the correct IQN. Watch out for MTU mismatches: a jumbo frame configured on the switch but not the array is a classic silent performance problem.
Host persona — a host object must be created with the correct persona (Windows, VMware, AIX, Linux), because it changes how LUN 0, REPORT LUNS and certain SCSI commands behave.

Multipath health is best verified from the host, not the array: on Linux use multipath -ll and look for the expected number of active paths, on VMware check the datastore's path state. A volume with only one active path will work perfectly until the moment it does not.

Health Monitoring and Day-2 Operations

Two commands form the core of routine operation, and both are cheap enough to run daily from a script:

cli% checkhealth
cli% shownode -state
cli% showpd -showcols Id,Cage,State,CurrentSize
cli% showalert -d
cli% showtask -active

checkhealth reports component states and flags anything that has degraded; treat a change in its output as an event worth investigating even if the array is still serving I/O. showalert lists outstanding alerts, and an unresolved alert on a drive or a port is the earliest warning you will get that a component is heading for replacement.

When a drive does need replacing, the workflow is entirely CLI-driven and must be followed in order — the array expects you to move the failed drive offline through the servicemag sequence rather than pulling it from the shelf. Skipping that step can leave the shelf in a degraded state that requires a cabled reset to clear.

Finally, keep capacity under review. On a thin-provisioned array it is easy to write far more data than the physical space allows, and reaching 100% of a CPG stops every write to volumes in that pool, not just the one that grew. Set alerts at 70% and 85% of usable capacity, and periodically check the ratio of provisioned to used space with showcpg -sdg.