RAID Capacity Calculator

Choose the RAID / mirror configuration, set how many disks there are (type it, or use - and +) and the capacity of one disk, or pick a common size and edit it. The result appears as you type. When you change units, the capacity stays the same and the input value is converted. Values that cannot be shown exactly are rounded for display, while calculations use the exact capacity.

Drive makers count 1 TB as 1,000,000,000,000 bytes. File Explorer in Windows commonly shows sizes converted with 1 KB = 1,024 bytes but labeled GB and TB, so the same capacity looks smaller there: 24 TB of usable space appears as about 21.8 TB. The "Shown in Windows" figure approximates that label; it is not the free space you will see after formatting, which is smaller again because of partitions, the file system and other reservations. The binary-unit value (TiB / GiB) is the same capacity in correctly named units.

RAID 0 stripes data with no redundancy: one failed disk loses everything. RAID 1 here is a mirror of two disks: the capacity of one disk, and either disk may fail. The multi-way mirror keeps the same data on three or more disks, as a RAID 1 over more devices does in Linux md: still one disk of capacity, and any N - 1 disks may fail. It is not a numbered standard RAID level. RAID 5 gives one disk's worth to parity and survives any one failure; RAID 6 gives two and survives any two. RAID 10 stripes over two-disk mirror pairs: it survives any one failure, and more only when every failure is in a different pair. Losing both disks of one pair loses the data.

Disks of different capacities go in further rows ("Add disks of another capacity"). Except in JBOD and RAID 50 / 60, every disk is then used only up to the smallest one, and the rest is shown as unusable. Linux md RAID 0 is an exception that can use more; the page says so. For RAID 50 / 60 each row is a span: enter every span's disk count and capacity, since the page never splits a total evenly. Each span counts as a full RAID 5 / 6 with its own parity, its disks used up to the smallest disk in that span (not the smallest of all spans), and the stripe over the spans uses each span up to the smallest span's usable capacity. If spans differ in disk count or capacity, the rest of a larger span is shown as unusable and the page flags the result as a theoretical estimate; Broadcom MegaRAID, for one, requires the same number of drives in every span. Dedicated hot spares (under "Advanced settings") count toward the installed capacity and disks, never toward the usable capacity, and do not raise the number of disks that may fail at once; the page checks that a spare holds at least the capacity each member uses, and RAID 0, a single disk and JBOD take none. That check is theoretical: whether a product accepts a disk as a spare is its own rule, which can involve the disk's real sector count, the capacity the controller adjusts members to, the interface and the media type (HDD or SSD). A spare at least as large as the largest member's nominal capacity is a common guide, not a universal requirement, and a calculated result does not guarantee that the configuration can be built. RAID 01 has RAID 10's capacity but one failure takes its whole stripe set out, so a second failure in the other set loses the data. RAID 1E keeps two copies of every block over three or more disks (half the capacity); beyond one failure it depends on the implementation's layout, which is not calculated. RAID 3 and RAID 4 put all parity on one disk, with RAID 5's capacity.

The capacity breakdown under the result splits the installed capacity into data, redundancy (mirror copies or parity, named after the level), hot spares and unusable space, each with its share. It shows shares of capacity, not where data sits: parity in RAID 5 / 6, for example, is spread over every disk, not kept on one. Each figure is rounded on its own, so when they do not add up exactly to the installed capacity or to 100% the page says so. "Compare RAID levels" calculates every level from the same disks and hot spares, one by one, with the same formulas as the result. A level that cannot take those disks is listed with the reason instead of being left out. RAID 50 / 60 are compared only when the disks are already entered as spans; the other levels then use all the disks without spans. Performance and rebuild time are not compared.

Dell ADAPT (ME4 8+2, ME5 8+2 or 16+2) and HPE MSA-DP+ (Gen6, Gen7) are listed after the standard levels as conditional estimates, one choice per product generation. They keep spare capacity on every disk (a distributed spare) instead of a dedicated hot spare, and stripe data and two parity over a fixed number of disks. The page assumes disks of one type and one capacity (it does not check the type), reserves the default spare (the two largest disks' worth) and splits the rest by the stripe width; metadata, formatting, rounding and system reserves are not subtracted, so a device can report less, and a calculated result does not mean a product can be built with those disks. Mixed capacities, disks of different types, dedicated hot spares, custom spare targets and groups above Dell's documented limits are not calculated. The result lists the documents it rests on. These schemes are not part of the level comparison, and a total of several groups that includes one says so.

NetApp ONTAP RAID-DP and RAID-TEC are given as theoretical values per RAID group: each row is one RAID group, and each group gives 2 (RAID-DP) or 3 (RAID-TEC) disks to parity, not the aggregate as a whole. Enter every group; the page never splits a total into groups. The documentation gives a minimum of 5 disks for RAID-DP and 7 for RAID-TEC. Values come from the nominal disk capacity, without ONTAP's right-sized capacity, partitioning, the root aggregate or WAFL and Snapshot reserves, and a result does not mean a system can build the configuration (model limits are in NetApp Hardware Universe). Spares are disks outside the groups and can be entered as hot spares.

IBM DRAID6 (Storage Virtualize 8.7) is a conditional estimate from the approximation in IBM Redbook SG24-8585. Choosing it asks for the stripe width (5 to 16) and the rebuild areas (1 to 4); the width plus the rebuild areas must not exceed the drives. The fields start at 12 and 1 in this calculator; the values an IBM system picks by itself depend on the drive count and the configuration, so enter your system's. The rebuild areas are distributed spare capacity. Parity is shown as 2/W of the other drives, and what the formula's correction term takes besides is shown as its own share, since IBM does not break it down. Mixed capacities, dedicated spares and the effective capacity of compressing drives (FCM) are not calculated; check the actual capacity on the system.

"Disk failure simulation" (collapsed under the result) lets you press disks to mark them as failed for RAID 0, RAID 1, the multi-way mirror, RAID 5, RAID 6 and RAID 10 (up to 64 disks), and says whether the array is healthy, running degraded or has lost its data, in text as well as color. RAID 10 is shown as mirror pairs of disks 1-2, 3-4 and so on, an assumed layout that a real controller may arrange differently: failures in different pairs keep the data, both disks of one pair lose it. It is a simulation on a mathematical model; rebuilds, hot spares taking over, rebuild time and failure rates are not modeled, and hot spares never add to the disks that may fail at once.

To plan separate RAIDs, such as an OS mirror and a data array, use "Add a group" under "Advanced settings". Each group is its own set of physical disks with its own level and disk size; you edit one group at a time, and the table lists every group with the total installed and usable capacity. An added group has no disk capacity yet and is shown as not set up until you enter one. Groups not set up and groups with input errors are left out of the total, which is then marked as partial. Removing a group asks for confirmation first. LUNs or partitions carved from one RAID are not modeled, and the name of a group does not assign a drive letter.

RAID keeps a system running through disk failures; it is not a backup. Other vendor-specific schemes and several RAIDs on the same disks are not calculated. "Vendor schemes: what this calculator covers" at the end of the page lists which vendor schemes are calculated, which are explained only (such as NetApp E-Series Dynamic Disk Pools) and which need another model (such as ZFS and vSAN), with the reasons and documents.