Skip to content

Dear customers! Please send orders and inquiries to [email protected] or to Telegram @servix_uz instead of calling by phone: due to the high workload, the operator cannot answer all incoming calls.

Servix

SAN, NAS or all-flash: how to choose a storage system for your company

Servix technical team · Updated

In brief. Virtualization clusters and databases need a block SAN array, file archives need NAS or a unified system, and all-flash NVMe is chosen for heavily loaded DBMS and VDI with sub-millisecond latency. For one to four servers and up to 100 TB, a SAS connection to an HPE MSA 1060 or Dell PowerVault ME5012 is enough; for a cluster of 4–⁠16 nodes, iSCSI 10/25GbE is suitable; Fibre Channel is justified only if there is an existing fabric. Two active/active controllers and a hot spare from day one are mandatory, and the 5:1 reduction ratio of PowerStore does not apply to video and archives.

Choosing a storage system almost never starts with terabytes. It starts with three questions: what workload the array will carry, how long the business can survive without access to the data, and where the system will grow in three years. The answers determine the architecture, the connection protocol and the drive type — and capacity turns out to be the last parameter to be calculated. Let’s look at how to choose a storage system on the merits, without marketing language.

Three architectures: DAS, SAN and NAS

DAS (Direct Attached Storage) is an enclosure connected directly to a server via SAS. No switches, no separate network, minimal latency and the lowest price per terabyte. The limitation is fundamental: the enclosure serves one or two servers, and if the server is off, the data is unavailable. A classic example of such an enclosure is the 84-drive Dell PowerVault ME484: a JBOD expansion without its own controllers that increases the capacity of an existing array.

SAN (Storage Area Network) is a block storage network. The array presents “raw” volumes to servers, which the operating system sees as local disks. This is the only option for virtualization clusters, databases and any scenario where several nodes must work with shared storage simultaneously: VMware vSphere HA, Hyper-V Failover Cluster, Oracle RAC. The vast majority of entry-level and midrange arrays — MSA, PowerVault ME5, PowerStore — are SAN.

NAS (Network Attached Storage) is file access over SMB and NFS. The array presents network shares rather than volumes; permissions are managed at the folder level and integrated with Active Directory. This is for file archives, departmental shares, video surveillance folders and backups, not for virtualization.

The boundary between SAN and NAS has long been blurred: unified systems serve both blocks and files from one array. Dell PowerStore, Huawei OceanStor and Unity XT can do this out of the box. Entry-level arrays — MSA and PowerVault ME5 — are block-only; file access on them is provided through a separate gateway server.

Connection protocol: SAS, iSCSI or Fibre Channel

This is where people overpay most often. Variants of the same array differ only in their interface modules, but the infrastructure around them costs different amounts.

Protocol

What needs to be purchased additionally

Who it suits

12G SAS

Nothing except cables and HBAs in the servers

1–⁠4 servers in one rack, minimal budget, the simplest possible setup

iSCSI 10/25GbE

A pair of Ethernet switches, preferably dedicated to storage traffic

Clusters of 4–⁠16 nodes, when the company already has Ethernet expertise

Fibre Channel 16/32G

FC switches, FC adapters in each server, separate zoning

Heavily loaded databases, predictable latency requirements, an existing FC fabric

A practical rule: if the company does not already have an FC fabric deployed, building one for a single array is usually not cost-effective — switches and adapters add a noticeable amount to the project, while the latency gain on typical virtualization workloads is measured in fractions of a millisecond. iSCSI at 10 or 25 gigabits covers the vast majority of midsize business tasks. The reverse situation: if FC is already in place and maintained, it makes sense to buy the new array in an FC version: the HPE MSA line offers the same platform in FC, iSCSI and SAS variants.

Hybrid or all-flash

A hybrid array combines SSDs for “hot” data with high-capacity SAS or NL-SAS drives for the rest, automatically moving blocks between tiers. All-flash runs only on solid-state drives and saves space through inline deduplication and compression.

The difference in how capacity is treated is fundamental. For a hybrid, usable capacity is calculated directly: drives minus RAID overhead. For all-flash, the manufacturer calculates effective capacity taking the data reduction ratio into account. Dell, for example, guarantees a reduction of 5:1 across the entire PowerStore line — from the entry-level 500T to the top models, without a prior workload assessment and without additional licenses. For virtual machines and databases, this ratio is achievable; for already compressed data — video, archives, encrypted files — it is not. Planning capacity for backups and media in effective terabytes is a typical project mistake.

What this means for specific models

Below are four classes of systems that are actually available in Tashkent, with specifications from the vendors’ official documents.

Model

Class

Supported drives

Connectivity

When to choose

HPE MSA 1060

Block, entry level, 2U

Up to 96 SFF or 36 LFF, up to three expansion enclosures

16G FC, 1/10G iSCSI, 12G SAS, 4 ports per array

1–⁠3 servers, a cluster of up to 8 nodes, a limited budget

Dell PowerVault ME5012

Block, entry level, 2U

Up to 264 drives with ME484 enclosures

FC, iSCSI, SAS — depending on the modules chosen

When you need maximum headroom in drive count within the same class

Huawei OceanStor 5500 V5

Hybrid, midrange

25 SFF in the base chassis, expandable with enclosures

FC, iSCSI, unified access

Mixed workloads, files and blocks from one system

Dell PowerStore 500T

All-flash NVMe, 2U

25 NVMe slots in the base, up to 97 per appliance

FC, iSCSI, NVMe/TCP, blocks and files

Databases and VDI, latency requirements, growth over years

Performance figures should be read as carefully as capacity. For the MSA 2060 , HPE claims up to 395,000 IOPS; for the PowerVault ME5, Dell claims up to 640,000 IOPS at an 8 KB block size and up to 12 GB/s of sequential read. These are peak synthetic values with a fully loaded cache and the maximum drive configuration. An array configured with twelve drives will show noticeably less, and that is normal: for a typical infrastructure of 30–⁠80 virtual machines, there will still be headroom.

Fault tolerance: what to consider mandatory

Two controllers in active/active mode are not an option but a basic requirement for any system serving production workloads. MSA, PowerVault ME5 and PowerStore all ship with two controllers: if one fails, the other takes over the volumes, and the servers do not notice. Single-controller configurations are acceptable only for backups and test environments.

Next come RAID levels. RAID 10 provides maximum write speed at the cost of half the capacity; RAID 5 saves space but is risky with large drives: rebuilding an array of 16–⁠20 TB drives takes a day or more, and all that time the array is vulnerable to a second drive failure. For high-capacity NL-SAS drives, a sensible choice is RAID 6 or its dual-parity equivalents; for flash pools, the vendor’s distributed schemes.

Spare drives are a separate matter. A hot spare should be in the configuration from day one: waiting for a drive to be shipped from abroad while the array runs without redundancy is a bad scenario.

Three things to check before buying in Uzbekistan

The model’s support status. Previous-generation equipment costs noticeably less on the market, but every line has an end-of-service-support date set by the vendor. Example: IBM Storwize V3700 is a reliable and still sought-after system, but IBM’s service support for it has ended: microcode updates and security patches are no longer released for it. Such an array makes sense for a specific task — expanding an existing fleet, setting up a test environment, covering an archive — but not as the main platform for the next five years. The same applies to the MSA 1050 of the previous generation relative to the current 1060/2060 line.

Drive compatibility. Enterprise-class arrays accept only drives with vendor firmware. A drive from the same manufacturer bought separately will most likely not be recognized by the array or will be flagged as unsupported. Capacity expansion is planned together with the purchase of the system, not as “we’ll buy more locally in a year.”

Delivery times for equipment and spare parts. In Uzbekistan, this is often more important than the specification. Ask where the equipment comes from, how long delivery takes, and whether the distributor’s warehouse has critical spare parts — controllers, power supplies, drives of the right model. An array that stands idle for a week waiting for a controller from abroad costs more than a more expensive system with a local warehouse.

A short selection algorithm

One or two servers, up to 100 TB, a limited budget — an entry-level SAS array or an expansion enclosure. A virtualization cluster of 4–⁠12 nodes with mixed workloads — a block array with iSCSI 10/25GbE, a hybrid configuration with a flash tier for virtual machines. Heavily loaded databases, sub-millisecond latency requirements, a growth plan for several years — all-flash NVMe with unified access. File archives and shared resources — NAS or a unified system, but not a block array with a server in between.

The full list of available systems is in the Storage systemssection: Dell PowerVault, PowerStore and Unity, HPE MSA and Alletra, Huawei OceanStor. If you need to turn your requirements into a specific configuration with a calculation of usable capacity for your workload, write to us, and we will select and calculate it.

Related collections and tools: Dell PowerVault ME5, HPE MSA 2060 and 2062, Dell PowerVault reference, HPE MSA reference.

Questions and answers

Can you start with one enclosure and expand later?

Yes, if you start with an array with controllers rather than a JBOD. Both entry-level platforms discussed support expansion enclosures: the MSA 1060 supports up to three additional enclosures, and the PowerVault ME5012 can be expanded to 264 drives. It is important to include the enclosure model in the project in advance — enclosures are not interchangeable between lines.

Which is more cost-effective: an array or local drives in the servers?

Local drives are cheaper as long as there are one or two servers and no cluster is needed. As soon as you need live migration of virtual machines, automatic restart on another node, or shared access by several servers to the same data, you cannot do without shared storage — either an external array or a software-defined solution on the servers themselves.

Is a separate array needed for backups?

Backups should not reside on the same system as production data — this violates the basic rule of backup. For backups, a separate high-capacity system or a specialized hardware deduplication appliance is usually used, and copies are additionally moved to media that cannot be written to from the production network.

Related reading