The world's first hard drive, the IBM RAMAC 305, which was released in 1956, contained only 5 MB of data, and weighed 970 kg and was comparable in size to an industrial refrigerator. Modern corporate flagships can boast a capacity of already 20 TB. Just imagine: 64 years ago, in order to record this amount of information, it would have taken more than 4 million RAMAC 305, and the size of the data center needed to accommodate them, would have exceeded 9 square kilometers, while today a small box weighing about 700 grams! In many ways, this incredible increase in storage density has been achieved thanks to the improvement of magnetic recording methods.
It's hard to believe, but fundamentally the design of hard drives has not changed for almost 40 years, since 1983: it was then that the first 3,5-inch hard drive RO351, developed by the Scottish company Rodime, saw the light. This baby received two magnetic plates of 10 MB each, that is, it was able to hold twice as much data as the updated 412-inch ST-5,25, released by Seagate in the same year for the IBM 5160 personal computers.
Rodime RO351 - the world's first 3,5-inch hard drive
Despite the innovativeness and compact size, at the time of the release of the RO351, almost no one needed it, and all further attempts by Rodime to gain a foothold in the hard drive market failed, which is why the company was forced to cease operations in 1991, having sold almost all existing assets and reducing state to a minimum. However, Rodime was not destined to go bankrupt: soon the largest hard drive manufacturers began to turn to her, wishing to acquire a license to use the form factor patented by the Scots. 3,5" is now the industry standard for both consumer and enterprise HDDs.
With the advent of neural networks, Deep Learning and the Internet of Things (IoT), the volume of data created by mankind has begun to grow like an avalanche. According to the estimates of the analytical agency IDC, by 2025 the amount of information generated by both the people themselves and the devices around us will reach 175 zettabytes (1 Zbyte = 1021 bytes), and this despite the fact that in 2019 it was 45 Zbytes, in 2016 - 16 Zbytes, and back in 2006, the total amount of data produced in the entire foreseeable history did not exceed 0,16 (!) Zbytes. Modern technologies help to cope with the information explosion, among which improved data recording methods are not the last.
LMR, PMR, CMR and TDMR: what's the difference?
The principle of operation of hard drives is quite simple. Thin metal plates coated with a layer of ferromagnetic material (a crystalline substance that can remain magnetized even in the absence of an external magnetic field at a temperature below the Curie point) move relative to the block of recording heads at high speed (5400 rpm or more). When an electric current is applied to the writing head, an alternating magnetic field arises, which changes the direction of the magnetization vector of the domains (discrete regions of matter) of the ferromagnet. Data reading occurs either due to the phenomenon of electromagnetic induction (the movement of domains relative to the sensor causes the occurrence of an alternating electric current in the latter), or due to the giant magnetoresistive effect (the electrical resistance of the sensor changes under the influence of a magnetic field), as is implemented in modern storage devices. Each domain encodes one bit of information, taking the logical value "0" or "1" depending on the direction of the magnetization vector.
For a long time, hard drives used the Longitudinal Magnetic Recording (LMR) method, in which the domain magnetization vector lay in the plane of the magnetic platter. Despite the relative ease of implementation, this technology had a significant drawback: in order to overcome coercivity (the transition of magnetic particles to a single-domain state), an impressive buffer zone (the so-called guard space) had to be left between tracks. As a result, the maximum recording density that was achieved at the end of this technology was only 150 Gb/in2.
In 2010, LMR was almost completely replaced by PMR (Perpendicular Magnetic Recording - perpendicular magnetic recording). The main difference between this technology and longitudinal magnetic recording is that the magnetic directivity vector of each domain is located at an angle of 90Β° to the surface of the magnetic plate, which made it possible to significantly reduce the gap between tracks.
Due to this, the data recording density has been significantly increased (up to 1 Tbit / inch2 in modern devices), while not sacrificing the speed characteristics and reliability of hard drives. At present, perpendicular magnetic recording is dominant in the market, which is why it is also often called CMR (Conventional Magnetic Recording - conventional magnetic recording). At the same time, one must understand that there is absolutely no difference between PMR and CMR - this is just a different version of the name.
When looking at the specifications of modern hard drives, you may also come across the cryptic abbreviation TDMR. In particular, this technology is used by enterprise-class drives . From the point of view of physics, TDMR (which stands for Two Dimensional Magnetic Recording - two-dimensional magnetic recording) is no different from the usual PMR: as before, we are dealing with non-intersecting tracks, domains in which are oriented perpendicular to the plane of the magnetic plates. The difference between technologies lies in the approach to reading information.
In the block of magnetic heads of hard drives created using TDMR technology, each recording head has two reading sensors that simultaneously read data from each passed track. This redundancy allows the HDD controller to effectively filter electromagnetic noise caused by Intertrack Interference (ITI).
Solving the problem with ITI provides two extremely important benefits:
- reduction of the noise factor allows to increase the recording density by reducing the distance between tracks, providing a gain in total capacity up to 10% compared to conventional PMR;
- Combined with RVS technology and a three-position micro actuator, TDMR effectively resists rotational vibration caused by hard drives, helping to achieve consistent levels of performance even in the most demanding environments.
What is SMR and what is it eaten with?
The dimensions of the writing head are about 1,7 times larger than the dimensions of the read sensor. Such an impressive difference is explained quite simply: if the recording module is made even more miniature, the strength of the magnetic field that it can generate will not be enough to magnetize the domains of the ferromagnetic layer, which means that the data simply will not be stored. In the case of a reading sensor, this problem does not arise. Moreover, its miniaturization makes it possible to further reduce the influence of the ITI mentioned above on the process of reading information.
This fact formed the basis of tiled magnetic recording (Shingled Magnetic Recording, SMR). Let's understand how it works. When using traditional PMR, the writing head moves relative to each previous track by a distance equal to its width + the width of the protective space (guard space).
When using the tiled method of magnetic recording, the recording head moves forward only a part of its width, so each previous track is partially overwritten by the next one: the magnetic tracks overlap each other like roof tiles. This approach makes it possible to further increase the recording density, providing a capacity gain of up to 10%, while not affecting the reading process. An example is β the world's first 3.5-inch 20 TB drives with SATA/SAS interface, the appearance of which was made possible thanks to the new magnetic recording technology. Thus, the transition to SMR disks allows you to increase the density of data storage in the same racks at minimal cost for upgrading the IT infrastructure.
Despite such a significant advantage, SMR has an obvious disadvantage. Since the magnetic tracks overlap each other, when updating data, it will be necessary to rewrite not only the required fragment, but also all subsequent tracks within the magnetic platter, the volume of which can exceed 2 terabytes, which is fraught with a serious drop in performance.
Combining a certain number of tracks into separate groups called zones helps to solve this problem. Although this approach to data storage somewhat reduces the overall capacity of the HDD (since sufficient gaps must be maintained between zones to prevent overwriting tracks from neighboring groups), this can significantly speed up the data update process, since now only a limited number of tracks participate in it.
Tiled magnetic recording involves several implementation options:
- Drive Managed SMR (Drive Managed SMR)
Its main advantage is that there is no need to modify the software and/or hardware of the host, since the HDD controller takes over the control of the data recording procedure. Such drives can be connected to any system that has the required interface (SATA or SAS), after which the drive will be immediately ready for use.
The disadvantage of this approach is performance variability, which makes Drive Managed SMR unsuitable for enterprise applications where system performance consistency is critical. However, such disks perform well in scenarios that allow sufficient time for background data defragmentation to complete. So, for example, DMSMR drives Optimized for use in small 8-bay NAS, it is an excellent choice for an archiving or backup system that requires long-term backup storage.
- Host Managed SMR (Host Managed SMR)
Host Managed SMR is the most preferred tile implementation for enterprise use. In this case, the host system itself is responsible for managing data flows and read / write operations, using for these purposes the extensions of the ATA (Zoned Device ATA Command Set, ZAC) and SCSI (Zoned Block Commands, ZBC) interfaces developed by the INCITS T10 and T13 committees .
When using HMSMR, the entire available storage capacity is divided into two types of zones: Conventional Zones (regular zones), which are used to store metadata and arbitrary recording (in fact, play the role of a cache), and Sequential Write Required Zones (sequential write zones), which occupy a large part of the total capacity of the hard disk, in which data is recorded strictly sequentially. Unordered data is stored in the cache area, from where it can then be transferred to the corresponding sequential write zone. Due to this, all physical sectors are written sequentially in the radial direction and are overwritten only after a wraparound, which allows you to achieve stable and predictable system performance. At the same time, HMSMR drives support random read commands similar to drives using standard PMR.
Host Managed SMR implemented in enterprise-class hard drives .
The line includes high-capacity SATA and SAS drives designed for use in hyperscale data centers. Support for Host Managed SMR significantly expands the scope of such hard drives: in addition to backup systems, they are perfect for cloud storage, CDN or streaming platforms. The high capacity of hard drives allows you to significantly increase storage density (in the same racks) with minimal upgrade costs, and low power consumption (less than 0,29 watts per terabyte of stored information) and heat dissipation (on average 5 Β° C lower than analogues) β further reduce the operating costs of maintaining the data center.
The only disadvantage of HMSMR is the comparative complexity of implementation. The thing is that today not a single operating system or application can work with such drives out of the box, which is why major changes in the software stack are required to adapt the IT infrastructure. First of all, this concerns, of course, the OS itself, which in the conditions of modern data centers using multi-core and multi-socket servers is a rather non-trivial task. You can learn more about options for implementing support for Host Managed SMR on a specialized resource. dedicated to the issues of zonal data storage. The information collected here will help you to preliminarily assess the readiness of your IT infrastructure for transition to zoned storage systems.
- Host Aware SMR (SMR supported by the host)
Host Aware SMR-enabled devices combine the convenience and flexibility of Drive Managed SMR with the fast recording speed of Host Managed SMR. Such drives are backward compatible with legacy storage systems and can function without direct control from the host, but in this case, as with DMSMR drives, their performance becomes unpredictable.
Like Host Managed SMR, Host Aware SMR uses two types of zones: Conventional Zones for random writes and Sequential Write Preferred Zones (zones preferred for sequential recording). The latter, in contrast to the Sequential Write Required Zones mentioned above, are automatically transferred to the category of ordinary ones if they begin to write data in an unordered manner.
The host-aware implementation of SMR provides internal mechanisms to recover from inconsistent writes. Random data is written to the cache area, from where the disk can transfer information to the sequential write zone after all the necessary blocks have been received. The drive uses an indirection table to manage inorder writes and background defragmentation. However, if predictable and optimized performance is required for enterprise applications, this can still be achieved only when the host takes full control of all data flows and write zones.
Source: habr.com