Disk and Storage Guide: Media, Interfaces, Protocols, Safe Inventory, and Migration

Table of Contents

Put each storage term back in its proper layer

“SSD, M.2, NVMe, PCIe, AHCI, RAID, and NTFS” answer different questions and do not belong in one comparison column. SSD is a media class, not an interface; M.2 is a form-factor and connector specification, not a synonym for NVMe; RAID is a data layout/controller presentation, not a backup; a filesystem sits above the block device.

This guide was last checked on September 1, 2026. It establishes a current taxonomy, then gives read-only, evidence-first Linux and Windows inventories and migration gates. The maintained section contains no command to format, erase, create an array, change controller mode, write a partition table, mount an unknown filesystem, or start a destructive test.

1. A six-layer model: one question per layer

Layer	Typical values	What it describes	What it cannot prove alone
Storage media	HDD, NAND-flash SSD, persistent memory	How data is physically retained	Connector, protocol, speed, or endurance
Form factor/connector	3.5-inch, 2.5-inch, M.2, U.2, EDSFF	Mechanical dimensions, keying, slot/cable shape	Actual bus, command set, or compatibility
Interconnect/transport	PATA, SATA, SAS, PCIe, USB	The link over which data moves	Whether ATA, SCSI, or NVMe is used above it
Command set/protocol	ATA, SCSI, NVMe	How a host requests read, write, flush, identify, and telemetry	Device shape or final OS path
Controller presentation	AHCI, HBA, hardware RAID, Intel RST/VMD, USB bridge	What device firmware/drivers expose to the OS	Number of physical media behind it or telemetry passthrough
Block/volume/filesystem	Namespace/LUN, partition, md/Storage Spaces, LVM, dm-crypt, NTFS/ext4/XFS	How the OS composes, encrypts, formats, and mounts blocks	Interchangeable underlying drives or an independent backup

Record layers top-to-bottom during diagnosis and verify them bottom-to-top during migration. An OS name such as /dev/sda, /dev/nvme0n1, or Disk 0 can represent a physical drive, namespace, RAID virtual disk, SAN LUN, or logical device behind a USB bridge; the name is not evidence by itself.

2. HDD and SSD are media, not interfaces

An HDD stores data on rotating platters with heads, so relevant evidence includes rotational speed, seek behavior, vibration, reallocated/pending sectors, and mechanical faults. An SSD uses non-volatile semiconductor memory, so evidence includes controller, NAND type, write amplification, spare area, temperature, media errors, estimated wear, and power-loss protection. Both can use SATA or SAS; SSDs also commonly use PCIe/NVMe.

“SSD is faster than HDD” is a common workload generalization, not an identity rule. Sequential throughput, random latency, queue depth, behavior near full capacity, sustained write, firmware, thermal throttling, and workload all matter. Do not identify a consumer SSD, enterprise NVMe drive, or virtual disk from ROTA=0 alone; correlate model, serial/WWN, transport, controller, and vendor data sheet.

Repeated scans, benchmarks, and long self-tests can add I/O and mechanical stress to a failing HDD. Freeze changes and preserve existing logs/topology first. If the data is unique or legally evidentiary, stop ordinary inventory and use a controlled imaging/forensics process.

3. Form factors: 2.5/3.5, M.2, U.2, and EDSFF

Form factor	Common media/link	Compatibility facts to verify
3.5-inch drive	Usually HDD, with specialized SSDs; SATA or SAS	Bay, power, connector, controller/backplane, cooling
2.5-inch drive	HDD or SSD; SATA, SAS, or U.2 PCIe/NVMe	Height, connector key, power, backplane signal support
M.2 module	Commonly SATA/AHCI or PCIe/NVMe; other functions also exist	Key, length, lanes, SATA/PCIe wiring, firmware support; fitting does not prove operation
U.2 drive	Commonly 2.5-inch PCIe/NVMe through the SFF-8639 connector family	Cable/backplane, PCIe lanes, hot-plug, management, power
EDSFF	Data-center E1.S/E1.L/E3.S/E3.L families	Exact SFF specification, dimensions, lanes, power/thermal envelope, platform qualified list

SNIA SFF specifications maintain relevant connector/form-factor documents. Purchasing and migration must use full part numbers and qualified lists for drive, cable, backplane, motherboard/controller—not only “M.2” or “2.5-inch.” M.2 SATA and M.2 NVMe can look similar while requiring different signals and firmware support.

4. Separate the interconnect from the command set

PATA is the historical parallel ATA link and generally carries ATA commands. SATA is serial ATA and generally still carries ATA commands; AHCI is a common SATA host-controller programming interface. SAS is Serial Attached SCSI transport and generally carries SCSI commands, with expanders for enterprise topologies. T10 develops SCSI storage-interface standards, while the SNIA SAS definition provides a terminology cross-check.

PCIe is a general-purpose interconnect. Local NVMe SSDs commonly use the NVMe command set over PCIe, but a PCIe device is not automatically NVMe; NVMe also defines fabrics transports. NVM Express specifications are authoritative for NVMe command sets, transports, and management.

USB is a host/device transport. A storage bridge may expose SCSI-shaped commands through Bulk-Only Transport or UAS, then translate them to ATA/NVMe inside an enclosure. A bridge can hide serials, WWNs, sector sizes, TRIM/UNMAP, SMART, or NVMe logs. Never infer internal media or command set from the USB cable shape.

Many SAS HBAs/backplanes can support SATA drives, with limits dependent on vendor, expander, cable, and firmware; a SATA controller generally cannot operate a SAS drive. Check controller and enclosure qualified lists instead of turning one-way compatibility into a universal claim.

5. AHCI, RAID, and RST/VMD are controller presentations

AHCI commonly exposes SATA devices individually to the OS. Hardware/firmware RAID may expose only one virtual disk while hiding member drives, serials, and telemetry. Software RAID, LVM, device-mapper, and Windows Storage Spaces add further OS layers.

Platform features such as Intel RST/VMD can place SATA/PCIe/NVMe devices behind vendor controllers and drivers. An OS-reported bus type or “SCSI disk” may reflect a driver model rather than a physical SCSI connector. Preserve PCI/controller IDs, drivers, firmware, UEFI mode, and management-tool version in the inventory.

Do not switch IDE compatibility, AHCI, RAID, RST, or VMD in UEFI merely to “make a disk appear.” The installed OS may lack the boot driver and the array metadata/boot path may depend on the current mode, causing a boot failure or a different logical-device view. Build a full backup and rollback plan and test the vendor migration procedure on matching staging hardware. This guide provides no generic mode-switch command.

6. PIO, DMA, UDMA, and PATA device 0/1 are history

PIO, single/multi-word DMA, and Ultra DMA are ATA/PATA-era host transfer modes. PIO involves the CPU; DMA/UDMA reduce CPU data movement and increase link rates. UDMA modes correspond to historical ATA33/66/100/133 labels. They are not current SATA, SAS, or NVMe performance tiers and cannot classify SSD media.

A PATA ribbon cable can attach two devices, historically called master/slave; clearer records use device 0/device 1 and cable-select/jumper configuration. This selects devices on one PATA channel and says nothing about authority, replication, or RAID. Primary/secondary channels, 40/80-conductor cables, jumpers, and BIOS detection order belong to legacy repair work.

For legacy recovery, photograph motherboard/controller labels, drive labels, jumpers, cable orientation, and BIOS screens, then work through a write blocker or clone. Do not use a 2011 Windows Device Manager path as a current SATA/NVMe diagnostic procedure.

7. SATA naming and performance boundaries

The SATA-IO naming guidelines call for “SATA 1.5Gb/s, SATA 3Gb/s, SATA 6Gb/s” transfer-rate wording and avoid “SATA II/SATA III,” which can imply an ambiguous feature set. SATA-IO specification access is the official route to current specifications.

These are link signaling rates, not guaranteed application payload throughput. Encoding, protocol overhead, queues, controller, media, filesystem, encryption, thermals, and workload constrain measurements. Link negotiation is commonly backward-compatible, but connector, power, port multiplier, hot-plug, and feature support still need separate verification.

Do not make “SATA-2 SSD” a media class or promise 6 GB/s from “SATA 6Gb/s.” Lowercase b denotes bits; 6 Gb/s is not 6 GB/s.

8. Filesystems and volumes sit above devices

A path may be physical drive → controller/bridge → namespace or LUN → partition table → RAID/storage pool → encryption/LVM → filesystem → mount point/share. Real order varies and can include multipath, thin provisioning, snapshots, and network storage.

Cloning a filesystem does not guarantee application-consistent data. Copying a block device does not automatically repair GPT, bootloaders, duplicate UUIDs, encryption keys, database journals, or cloud attachments. Linux device-mapper documentation illustrates block mapping layers. Every migration needs a dependency graph identifying which layer provides redundancy, snapshots, encryption, and mounts.

When md, dm-*, LVM, multipath, ZFS, Storage Spaces, hardware RAID, or a SAN LUN appears, do not bypass the upper layer to manipulate a member. Use the stack's official read-only query tools and recovery documentation first.

9. Linux: read-only inventory with explicit fields

The `lsblk` official util-linux documentation warns that default output can change, so scripts must name columns explicitly. Let device events settle, then run these read-only queries. Results contain serial, WWN, mount, UUID, and internal paths; keep them in a restricted local evidence directory, not a public issue.

set -eu

OUT=storage-inventory-20260901T132000Z

test ! -e "$OUT"
install -d -m 0700 "$OUT"
uname -a > "$OUT/uname.txt"
lsblk --json --bytes --output NAME,KNAME,PATH,TYPE,SIZE,MODEL,SERIAL,WWN,TRAN,ROTA,LOG-SEC,PHY-SEC,DISC-GRAN,DISC-MAX,FSTYPE,FSVER,LABEL,UUID,MOUNTPOINTS > "$OUT/lsblk.json"
findmnt --json --real --output TARGET,SOURCE,FSTYPE,OPTIONS > "$OUT/findmnt.json"
cat /proc/mdstat > "$OUT/mdstat.txt"

if command -v lspci >/dev/null; then
  lspci -nnk > "$OUT/lspci.txt"
fi

if test -d /dev/disk/by-id; then
  ls -l /dev/disk/by-id > "$OUT/by-id.txt"
fi

The `findmnt` official documentation supports checking mount sources and options. Never identify source/target by /dev/sdX letter order alone; reboot, hot-plug, or controller changes can rename devices. Cross-check at least serial, WWN, model, capacity, by-id path, physical bay/slot, and controller path. Stop on any conflicting field.

10. SMART and NVMe telemetry are evidence, not guarantees

SMART attribute numbers, raw values, thresholds, and normalization are often vendor-defined. PASSED does not prove a drive will not fail, and a raw count is not comparable across models. A USB/RAID bridge may block passthrough, a query may spin up a standby HDD, and reading logs can worsen an already damaged device.

Use smartmontools documentation and the `smartctl` manual for option semantics. Review scan output and controller documentation before replacing placeholders with devices confirmed by serial/WWN. A read-only baseline does not start a self-test or change SCT/ERC, power, or firmware settings.

sudo smartctl --scan-open
sudo smartctl --xall --nocheck=standby /dev/sdX
sudo smartctl --xall /dev/nvme0

sudo nvme list --output-format=json
sudo nvme id-ctrl /dev/nvme0 --output-format=json
sudo nvme id-ns /dev/nvme0n1 --output-format=json
sudo nvme smart-log /dev/nvme0 --output-format=json

The NVMe health log has controller/namespace semantics. Temperature, available spare, percentage used, media/data-integrity errors, and unsafe shutdowns need vendor definitions, critical warnings, firmware, and workload context. The nvme-cli official project and Linux NVMe documentation define command/kernel boundaries. Record missing telemetry as “not visible,” never “healthy.”

11. Windows: read-only Storage cmdlet inventory

The Microsoft Storage PowerShell module distinguishes disks, physical disks, storage pools, virtual disks, partitions, and volumes. These cmdlets only query and save JSON. Their private identifiers also require restricted access.

$stamp = Get-Date -Format yyyyMMddTHHmmss
$out = "storage-inventory-$stamp"

if (Test-Path $out) { throw "Inventory directory already exists" }
New-Item -ItemType Directory -Path $out | Out-Null

Get-Disk | ConvertTo-Json -Depth 6 | Set-Content "$out/disks.json" -Encoding utf8
Get-PhysicalDisk | ConvertTo-Json -Depth 6 | Set-Content "$out/physical-disks.json" -Encoding utf8
Get-StoragePool | ConvertTo-Json -Depth 6 | Set-Content "$out/storage-pools.json" -Encoding utf8
Get-VirtualDisk | ConvertTo-Json -Depth 6 | Set-Content "$out/virtual-disks.json" -Encoding utf8
Get-Partition | ConvertTo-Json -Depth 6 | Set-Content "$out/partitions.json" -Encoding utf8
Get-Volume | ConvertTo-Json -Depth 6 | Set-Content "$out/volumes.json" -Encoding utf8
Get-CimInstance Win32_DiskDrive | ConvertTo-Json -Depth 6 | Set-Content "$out/win32-disk-drives.json" -Encoding utf8

Do not add Initialize-Disk, Clear-Disk, New-Partition, Format-Volume, Reset-PhysicalDisk, pool/virtual-disk repair, or controller-mode changes to the same diagnosis. If Get-Disk and Get-PhysicalDisk counts differ, inspect RAID, Storage Spaces, VHD, USB bridges, SAN, and multipath rather than guessing that a drive is missing.

12. Logical/physical sectors, alignment, and TRIM caveats

A device may report 512-byte logical sectors, 4096-byte physical sectors with 512-byte emulation (512e), or 4096-byte logical/physical sectors (4Kn). Partition alignment, boot stack, filesystem, hypervisor, enclosure, and cloning tool must support the source/target combination. Equal capacity does not prove logical-sector compatibility; a block clone to a different logical size may fail or not boot.

Linux LOG-SEC/PHY-SEC and Windows disk properties are observation points, not vendor guarantees. A bridge/RAID controller can rewrite reported values, and existing partition offsets must be checked. Never change sector format online. NVMe format, SCSI format, and vendor sector conversion can destroy all data.

TRIM, UNMAP, or discard tells SSD/thin storage which logical ranges are unused. It may improve space reclamation while reducing deleted-data recovery. Nonzero DISC-GRAN/DISC-MAX means one layer reports capability, not that encryption, RAID, USB, or SAN passes it safely end-to-end. The util-linux `fstrim` documentation describes discard; do not run it during evidence collection or on an unknown/recovery volume.

13. RAID provides availability, not an independent backup

A RAID level defines a live layout such as striping, mirroring, or parity. It can preserve service through certain member failures, but deletion, ransomware, filesystem corruption, bad firmware, a mistaken rebuild, controller loss, fire, and operator error can still affect the entire array. Because writes and deletions propagate to live members, RAID does not meet the definition of an independent recovery copy.

Backups should be separated from the live array, versioned, access-restricted, and include offline/immutable copies and regular restore tests. Inventory RAID implementation, level, member serial/slot, stripe/layout, metadata version, cache/BBU, firmware, spares, and rebuild state. Never pull a drive based only on /dev/sdX order or run assemble, create, initialize, clear-foreign, rebuild, or consistency repair on an unknown array.

When a controller exposes only a virtual disk, missing OS SMART does not prove healthy members. Export physical-drive and event logs with the controller vendor's read-only tools; first check that the command cannot acknowledge, clear, or start a patrol/read test.

14. Evidence-first migration gates

Phase	Must preserve/prove	Stop condition
Discovery	Layer graph, serial/WWN, slot, controller, sector size, volume/filesystem, encryption, boot role	Ambiguous source/target identity or unknown RAID/multipath/encryption
Protection	Application-consistent backup, files/database at one recovery point, keys, checksums, restore test	Only copy, untested backup, or rapidly degrading source
Design	Reason for file restore vs block image, capacity/sector/alignment/boot compatibility, downtime	Compatibility assumed from connector alone
Rehearsal	Isolated clone/staging, read-only source, test commands, expected hashes/counts, rollback	Test would write to the only source
Cutover	Freeze window, final increment, explicit source/target, service validation, monitoring	Identifiers conflict with approval record
Retention	Old source offline and unchanged, acceptance window, encrypted evidence, disposal approval	New system unverified or rollback window open

Prefer filesystem/application-aware restore when possible; it handles different capacities, sector sizes, and layouts better. Block imaging is appropriate when complete device state matters, but the target must satisfy size and sector/boot constraints. Connecting source and clone simultaneously can also create duplicate filesystem UUID, LVM ID, or array-metadata conflicts.

Stop ordinary migration when the source has I/O errors, abnormal sounds, rapidly increasing media errors, or the only copy of critical data. Do not benchmark, repair the filesystem, start a long self-test, power-cycle repeatedly, or auto-retry. Record current state and use a professional, retry-aware imaging workflow with an error map.

15. Validation and handoff checklist

[ ] Every term is assigned to media, form factor, transport, command set, controller, or volume/filesystem.
[ ] Source and target are cross-checked with two persistent identifiers plus physical slot.
[ ] Controller/driver/firmware, sector sizes, TRIM capability, and the full layer graph are preserved.
[ ] Raw SMART/NVMe/controller telemetry is retained with invisible fields documented.
[ ] SMART “passed,” RAID, and snapshots are not treated as backups.
[ ] Backups are encrypted, isolated, verified, and restore-tested in staging.
[ ] File-level or block-level migration has a written rationale and compatibility evidence.
[ ] Application, database, filesystem, and boot validation have explicit pass/fail criteria.
[ ] Writes are frozen before cutover; rollback artifacts and old source stay unchanged through observation.
[ ] Only after acceptance and retention expiry does an authorized person plan separate sanitization/destruction.

16. Current official and first-party references

17. Archive of the 2011/2023 source

The complete visible source_export body follows. One trailing ASCII space was removed from the old long table line for repository whitespace checks. To prevent WordPress from consuming a character while translations are saved in sequence, the same line's one historical backslash was replaced with the explicit marker [U+005C BACKSLASH]. No other text, link, or whitespace changed; only the inert outer code fence was added. The original export and Git history remain unchanged.

Warning: the archive mixes media, interfaces, and controller modes and contains obsolete speed claims, Windows paths, and operating advice. It is for historical audit only; use the maintained guide above.


磁盘相关知识的介绍

这个表格是我自己整理的，是我的体会，不一定准确

硬盘 接口 工作模式 备注 IDE硬盘 ATA ATA 理论上我们现在的将控制器做在电路板中，与硬盘合在一起的这种盘，都叫做IDE硬盘，这是IDE真正的含义 SATA硬盘 SATA IDEAHCIRAID 在Thinkpad T400 设置 IDE模式：**Config–>SATA–>Compatibility[兼容]** SCSI硬盘 SCSI SCSI 我在VMware中见过，安装Red hat Linux 的时候，要删除SCSI硬盘，新建IDE硬盘再安装 SSD硬盘 SATA-2 [U+005C BACKSLASH] SSD固定硬盘，我猜想和SATA差不多吧

Table of Contents

Toggle

- [IDE(ATA/PATA)、SATA、SCSI、SAS](https://blog.lazying.art/en/html/computer_internet/storage_disk/414/%e7%a3%81%e7%9b%98%e7%9b%b8%e5%85%b3%e7%9f%a5%e8%af%86%e7%9a%84%e4%bb%8b%e7%bb%8d.html/#IDEATAPATA%E3%80%81SATA%E3%80%81SCSI%E3%80%81SAS)
- [固态硬盘SSD](https://blog.lazying.art/en/html/computer_internet/storage_disk/414/%e7%a3%81%e7%9b%98%e7%9b%b8%e5%85%b3%e7%9f%a5%e8%af%86%e7%9a%84%e4%bb%8b%e7%bb%8d.html/#%E5%9B%BA%E6%80%81%E7%A1%AC%E7%9B%98SSD)
- [ATA传输模式：PIO、DMA、UltraDMA(UDMA)](https://blog.lazying.art/en/html/computer_internet/storage_disk/414/%e7%a3%81%e7%9b%98%e7%9b%b8%e5%85%b3%e7%9f%a5%e8%af%86%e7%9a%84%e4%bb%8b%e7%bb%8d.html/#ATA%E4%BC%A0%E8%BE%93%E6%A8%A1%E5%BC%8F%EF%BC%9APIO%E3%80%81DMA%E3%80%81UltraDMAUDMA)
- [主要/次要IDE通道、主/从盘](https://blog.lazying.art/en/html/computer_internet/storage_disk/414/%e7%a3%81%e7%9b%98%e7%9b%b8%e5%85%b3%e7%9f%a5%e8%af%86%e7%9a%84%e4%bb%8b%e7%bb%8d.html/#%E4%B8%BB%E8%A6%81%E6%AC%A1%E8%A6%81IDE%E9%80%9A%E9%81%93%E3%80%81%E4%B8%BB%E4%BB%8E%E7%9B%98)
- [SATA硬盘工作模式:IDE、AHCI和RAID](https://blog.lazying.art/en/html/computer_internet/storage_disk/414/%e7%a3%81%e7%9b%98%e7%9b%b8%e5%85%b3%e7%9f%a5%e8%af%86%e7%9a%84%e4%bb%8b%e7%bb%8d.html/#SATA%E7%A1%AC%E7%9B%98%E5%B7%A5%E4%BD%9C%E6%A8%A1%E5%BC%8F_IDE%E3%80%81AHCI%E5%92%8CRAID)
- [S.M.A.R.T](https://blog.lazying.art/en/html/computer_internet/storage_disk/414/%e7%a3%81%e7%9b%98%e7%9b%b8%e5%85%b3%e7%9f%a5%e8%af%86%e7%9a%84%e4%bb%8b%e7%bb%8d.html/#SMART)

## IDE(ATA/PATA)、SATA、SCSI、SAS

早期硬盘与控制器是分离的。

IDE（Integrated Drive Electronics），是指把“硬盘控制器”与“盘体”集成在一起的硬盘驱动器，也就是理论上我们现在的将控制器做在电路板中，与硬盘合在一起的这种盘，都叫做IDE硬盘，这是IDE真正的含义。

ATA（Advanced Technology Attachment），可能因为它是第一个IDE接口标准，现在基本都用IDE来指代ATA。平时可以将IDE与ATA理解为相同概念。

SATA（Serial ATA），串行ATA。

SCSI（Small Computer System Interface），SCSI接口。

SAS（Serial Attached SCSI），串行连接SCSI接口。

以上4种接口中，ATA与SCSI为并行方式传输数据，因此数据线针数较多。SATA与SAS为串行方式传输数据，数据线针数较少。

通常，SATA优于ATA，SAS优于SCSI。SCSI与SATA相比，具有更高的强度及稳定性，适合企业级服务器级的应用。SATA与SCSI相比，优势是比较便宜，实用。

## 固态硬盘SSD

solid state disk（固态硬盘），即用固态电子存储芯片阵列制成的硬盘，由控制单元和存储单元（DRAM或FLASH芯片）两部分组成。存储单元负责存储数据，控制单元负责读取、写入数据。

新一代的固态硬盘普遍采用SATA-2接口。

## ATA传输模式：PIO、DMA、UltraDMA(UDMA)

ATA早期有PIO和DMA两种模式，后来又有了现在的Ultra DMA模式。

PIO（Programmed I/O）是一种通过CPU执行I/O端口指令来进行数据的读写的数据交换模式，是最早期的一种传输方式。由于需要CPU干预，会非常占用CPU资源。并且速度很非常慢，有PIO mode 0（3.3MB/s） 至 PIO mode 4（16.6MB/s）4种模式。

DMA（Direct Memory Access）是一种不经过CPU而直接从内存存取数据的数据交换模式。有Single-Word DMA（8.33MB/s）和Multi-Word DMA（16.66MB/s）两种模式。

UDMA（Ultra DMA）目前有UDMA mod 2(ATA33)、UDMA mod 4(ATA66)、UDMA mod 5(ATA100)、UDMA mod 6(ATA133)，速度分别为33MB/s、66MB/s、100MB/s、133MB/s。

查看服务器上某块硬盘的传输模式的方法为:

设备管理器—IDE ATA/ATAPI控制器—主要/次要IDE通道—属性—高级设置—当前传送模式

现在的硬盘一般情况下运行在UDMA mod 5模式下，速度为100MB/s。UDMA6为拓迈公司提出的标准，尚未得到业界广泛支持，例如Inter目前只支持UDMA5。

如果硬盘有坏道，读取出现问题，系统的IDE/ATAPI 控制器的驱动程序(Atapi.sys)在累计收到6次超时或者是循环冗余校验(CRC)错误后，操作系统会自动将传输模式降低一级，如果今后持续收到超时或者CRC错误，最终传输模式将会降低到最地的PIO模式。此时硬盘传输速度将会非常慢。

## 主要/次要IDE通道、主/从盘

主板上的一个IDE插槽对应系统的设备管理器中的一个IDE通道。如果主板上有两个IDE插槽，一般会在主板上标记IDE1和IDE2，IDE1的为主要IDE通道，IDE2的为次要IDE通道。由于IDE逐渐被SATA淘汰，现在的主板大多只有一个IDE插槽。

每个IDE插槽上可以接两个IDE设备，一个主设备和一个从设备，不能两个同时为主或同时为从。主/从可通过设备的跳线设置。一般数据线的中间端接从盘，边缘端接主盘。

启动时检测的顺序为:先主要IDE通道，后次要IDE通道，先主盘，后从盘。

设备管理器—IDE ATA/ATAPI控制器—主/次要IDE通道—属性—高级设置—设备0/1—设备类型为”自动检测”表示处于开启状态，如果想关闭，可以设置成”无”，这样系统启动时不会检测，可加快启动速度，但如果有硬盘接在该位置，因为系统不检测，它也不会被识别。

## SATA硬盘工作模式:IDE、AHCI和RAID

SATA硬盘可以有IDE、AHCI和RAID三种工作模式

IDE模式:将SATA硬盘映射成IDE模式，这种方式不需要安装驱动，主板默认将SATA设置为该模式。SATA控制器在系统中也识别为主/次要IDE通道。

AHCI模式:该模式下的SATA硬盘将采用原生方式工作以支持NCQ（Native Command Queue，开启它可以提高硬盘性能）。AHCI需要主板支持，并且需要安装驱动（不装驱动直接在BIOS里修改成AHCI会导致启动时蓝屏无法进入系统）。使用该模式后启动时自检时间会变长，可以使用NCQ功能，支持热插拔。

RAID模式:多块硬盘做阵列。这时设备管理器中会多出”SCSI RAID控制器”一项（当使用了SCSI盘，或做了RAID，或安装了虚拟光驱，都会多出这一项）。

开启AHCI模式后，硬盘性能提升与默认的IDE模式相比并不是非常大。一般使用默认的IDE模式就可以。

## S.M.A.R.T

S.M.A.R.T的全称为“Self-Monitoring Analysis and Reporting Technology”，即“自我监测、分析及报告技术”。支持S.M.A.R.T技术的硬盘可以通过硬盘上的监测指令和主机上的监测软件对磁头、盘片、马达、电路的运行情况、历史记录及预设的安全值进行分析、比较。当出现安全值范围以外的情况时，就会自动向用户发出警告。

HDTune软件中的”健康状况”一栏，即显示了硬盘的S.M.A.R.T信息。

SATA硬盘在AHCI模式下，检测软件无法读取硬盘的S.M.A.R.T信息。