Why is RAID SSD much faster

SSD in server operation

The use of solid state drives (SSD), such as the Samsung SM863 or the PM863 is now also part of everyday life in the server environment. In this article, together with co-author Michael Elbel, IT manager at the Munich IT service provider ConSol, we illustrate the real strengths of flash storage in the data center compared to traditional hard drives. Using best practices, we show classic application examples and provide strategies on how you can optimally use the potential of SSDs in your individual corporate IT infrastructure.

The article at a glance:

SSDs: No loss of security in the server

Doubts about the suitability of flash memories in the server environment due to their lifespan and reliability are no longer justified today. The manufacturers of server SSDs have worked flat out over the past few years to significantly increase the number of guaranteed read and write accesses so that the lifespan of SSD and HDD in the professional sector has largely approximated. A 256GB SSD for server use actually contains between 300 and 350GB, depending on the manufacturer - even if the logic of the disk leaves out the buffer and consistently displays 256GB. If a block actually breaks, which is usually noticed when writing, the next block is written as a substitute.

In addition, “wear leveling” ensures that the available memory cells are written to in a balanced manner. For example, an SSD does not pack the Windows swap file (pagefile.sys) into the same memory cell every time it is saved, but rather distributes it over all writable areas. In this way, the failure of individual cells due to overload (on average after around 10,000 full write accesses) is significantly delayed.

In addition, in the event of a failure, aids are definitely available to get lost data out of a defective SSD, even if that is the issue Data recovery of SSDs is of course just as young as the technology itself.

In order to be prepared against the possible failure of a single disk, hard disks in servers - regardless of whether it is SSD or HDD - should be in RAIDs ("Redundant Array of Independent Disks"). Thanks to the redundantly generated information in the RAID system, data availability is guaranteed even if a single RAID component fails.

However, redundancy in RAID must not be confused with data security: it only protects against the failure of a disk in the RAID array, but not against the - in proportionately much more likely - data loss due to software or user errors. In front The decision to use an SSD or HDD as the storage medium within the server and the construction of a RAID system must be based on the careful elaboration of a backup concept - if not available - that not only clearly defines the data backup itself, but also includes organizational aspects such as For example, the storage of the backups spatially separated from the server for data retention in the event of fire and water damage.

In short: Anyone who decides to use SSDs in their server on the basis of a sensible backup concept and organizes them in a RAID does not have to accept any loss of security compared to classic hard drives.

Strengths of SSDs in server operation

While SSDs score points in mobile devices due to their low energy consumption and their special insensitivity to impacts, these aspects do not play a role in the data center. In server operation, the form factors of classic hard disks have changed drastically in the last few decades: apart from the technical development that ever larger storage capacities are being accommodated on ever smaller disks, there has been an increasing tendency to use smaller disks in recent years, also for reasons of space . Newer servers are therefore not necessarily equipped with 3.5 ", but often with 2.5" hard drives. The smaller the physical dimensions of a hard drive, the lower its power consumption. Measured by their capacities, the energy balances of HDD and SSD no longer differ too significantly - the CPU of a high-performance server often needs more power than the installed storage disks.

Rapid access - the real strength of flash memory

The real strength of SSDs lies in their performance. In the server environment, the data transfer rate (including data transfer rate) is less important, as it is over 200 MB / sec with current hard disks - regardless of whether they are HDD or SSD. beyond what is usually necessary in server operation. Often there are other limits to the data transfer speed, e.g. if the data is stored in a SAN (Storage Area Network).

It is completely different with the access time. This is significantly shorter with SSDs than with HDDs, in which the disk must first rotate and the head must move mechanically to the correct position. Since servers that run several applications are usually equipped with a large main memory (RAM), they do not have to access the hard disk for every single action. This means that the use of an SSD does not necessarily have to have a major impact compared to an HDD.

Drastic differences, on the other hand, can be measured in I / O-intensive applications in which, for example, several users work on the server at the same time and access data. While a classic hard drive offers 50-100 IOPS (I / O operations per second), an SSD with 10,000 IOPS provides a factor of 1,000 in performance plus. Typical areas of application for SSDs in servers can therefore always be found in the environment of applications that access, process and make available large amounts of data within a very short time - such as databases (MySQL, MS-SQL, Oracle ...) or Exchange servers.

Classic use cases and best practices for SSDs in servers

Example: 100 users receive 100 e-mails every hour and access their mailboxes continuously. It takes 10 milliseconds for each access until the desired data point is reached on the HDD. With 100 accesses per second, it is in this case without the use of SSDs on the Mail server fast tight.

Anyone who has an access-intensive application that is subject to performance restrictions should first investigate why it requires so many accesses / second. If everything is in order on the software side, the targeted replacement of the affected HDDs in the server with SSDs solves the bottleneck in such a case and brings enormous profit. This is also the case with the following ConSol customer project:
A large automobile manufacturer uses the ConSol * CM BPM software for lead management. In one Data warehouse All process-relevant information is continuously saved and made available for analysis. A report was to be drawn over the entire database once an hour - but initially it took a long time before the result was available. Background: The integrated analysis tool had to access an almost infinite number of data records for each report in order to compile a meaningful result. Today the entire application including data warehouse and reporting runs on server SSDs. The exchange meant that the data analysis was carried out five times faster than before on the classic hard drives in one fell swoop. The reports are now available for further use correspondingly faster.

An indication of the actual effect of the - usually partial - conversion of a server to SSD is provided by the Consideration of the working set. The working set is the amount of data within an application that is used frequently in a given period of time: For example, a merchandise management system includes all invoices and account movements over the past ten years. Actually, however, only the data from the last three months is normally used. While all data taken together have an estimated storage requirement of 20-200GB, in this case only around 2GB is regularly accessed; an order of magnitude that the main memory easily covers: it reads the required data once and only in exceptional cases does it have to access the disk again. The particular strength of an SSD would not have a significant impact here as long as the server is equipped with sufficient RAM capacity.

Tiering: Distribute data storage sensibly

In order to optimize storage capacity and IOPS at the same time, the hardware RAID controllers of the common server manufacturers offer the possibility of adding SSDs to existing HDDs in the storage subsystem, which then serve internally as intermediate storage and ensure that frequently used data is available as quickly as possible. In this way, the access speed can often be accelerated by a factor of 2 in one fell swoop, although HDDs are still mainly in use.

Compared to simple computers or even notebooks, servers basically offer a significant plus in flexibility due to their larger number of slots for equipping with SSD or HDD. Since different hard drives and capacities have very large price differences, one usually works with "tiering" when assembling servers. The entire data memory is divided into different groups and assigned to suitable storage media. Here is an example of a 3-stage tiering:

    • SSDs for a manageable data area that has to be accessed extremely quickly
    • SAS hard disks for larger areas with important and high-performance data that require fast access
    • Large and cheaper SATA hard drives for the data area that is rarely accessed (e.g. archiving)

Step by step: Targeted use of SSDs in the server infrastructure

So if you are faced with decreasing reaction times of your server and want to get the maximum performance, data security and availability out of your existing hardware, you should work out your individual storage target configuration step by step. First and foremost, there should be a sound data security concept with a clean backup.

1.       Know the causes

First of all, it must be examined in detail what has slowed the response time of the server: Has the number of users or the amount of data increased? Are new applications in use? Such a root cause analysis helps, on the one hand, to track down technical problems independently of storage bottlenecks and, if necessary, to rectify them; on the other hand, the results serve as a sound starting point for the following steps.

2.       Check options (1): upgrade RAM

If the system is to access a manageable working set more quickly, upgrading the main memory to maximum values ​​can quickly and easily ensure a noticeable increase in performance when accessing data.

3.       Check options (2): Check hard disk slots and connections

If the RAM upgrade is not enough or if it makes sense to equip the server with SSDs in view of future requirements, carry out a detailed inventory of your server: How many hard disk slots are there? How many of them are free? Which hard drives are currently installed (capacities, connections)?

A mounting frame is usually necessary for servers with 3.5-inch racks. Common adapter frames usually do not fit into hot-swap slots because the SATA connector is not in the same place as that of a 3.5-inch hard drive. If an existing 3.5-inch HDD is to be replaced by a hot-swap-capable SSD, installing it may require a bit of tinkering, which in turn makes the hot-swap option more difficult. This problem does not exist with 2.5-inch bays.

4.       Select SSDs

The amount of data that is to be stored on it in the long term is decisive for the selection of the right SSD capacity. In principle, any hard drive can be replaced by a faster storage medium with the same capacity. In most cases, however, larger hard disks are replaced by SSDs with a smaller capacity for cost reasons, so that data must be shifted before the migration (see point 6). In this case, a carefully developed tiering concept provides information about the appropriate SSD size.

5.       Design a RAID network

In order to be protected against the failure of an SSD, a RAID system should be planned for when upgrading. In particular, if the SSDs are part of a multi-level tiering concept, data mirroring as part of a simple RAID 1 with an SSD pair is usually sufficient. How you set up the RAID depends heavily on the type of server and the RAID controller used. As a rule, the network can be set up during operation.

If there is no longer enough space in the server for 2 - or possibly 4 - SSDs, free slots can be created using a combination of approaches. So it can be considered to first replace the existing hard drives with larger ones. Example: A server is equipped with four 500GB hard disks - all hard disk slots are occupied. In order to be able to provide a higher access speed with two SSDs, two terabyte hard drives are used and equipped with the contents of the 500GB HDDs, so that the storage capacity of the hard drives is basically retained and two slots are free at the same time.

6.       Data migration

Given the shortage of slots, a bit of creativity may be required for data migration while the system is running. In such cases, it is advisable to use external storage media, for example, in order to create some space to move the data. When planning the capacity, a clear concept should be in place as to which data should finally find its place on which disk. In this way, not only can the required capacities be determined in a targeted manner, but unwanted data loss and redundancies in the course of the migration can also be excluded from the outset. Basically, the actual migration steps depend heavily on the raid controllers and the operating system that are used.

(Update 3.2.2016) You can read about the specific lifespan of an SSD and how and whether this can be influenced in our article "The lifespan of an SSD - how long does it last and what can I do for it?"

Michael Elbelis IT manager at the Munich IT service provider ConSol and is responsible for the architecture, conception, implementation and operation of all IT infrastructure - in the server area from dedicated individual computers with special tasks (e.g. reporting server) to the company-wide VMware cluster with several TB of RAM and data storage in one multiple redundant, configured SAN. In its professional career, Elbel has seen and experienced the development of server infrastructures and the associated storage systems over the past 25 years - from washing machine-sized eight-inch HDDs with a few hundred MB capacity to high-performance SSDs currently connected directly via PCI Express actively supported as a system administrator and architect on a wide variety of operating systems.

You will find powerful Samsung SSDs for different areas of application at CompuRAM.
These Samsung SSDs are recommended for use in servers.

23.01.14, Updated on 07/23/20

Recommend this article:

Keywords: purchase advice, server, SSD