SMR technology opens up new horizons for magnetic recording. Tiled Magnetic Recording aka SMR

Modern home computers are equipped with both fast solid-state drives (SSD), with the speed of which no mechanical HDD can compete, and conventional HDDs, which are incomparably cheaper in price.

A kind of specialization has taken place: expensive and fast SSDs are used to run programs, and capacious and slow HDDs are used to store data. Therefore, the battle front of mechanical disks is the increase in capacity due to data compaction.

In pursuit of density, in 2007 Seagate proposed perpendicular recording technology. There was a lot of criticism about this at the time. But now this way of organizing data has become the standard, and without it it is not possible to create large disks.

Now Seagate has proposed a new technology - tiled recording.

Basic working principle

The term "shingled recording" comes from the English word - shingled magnetic recording (SMR). The technology helps to compress data so that the capacity of the drive becomes a quarter larger. In the classical disk surface layout scheme, all data is written to tracks that run along the surface one after another.

When using SMR, the tracks are also arranged one above the other. It turns out such a kind of partially third dimension in the placement of data.

The paths only partially overlap each other, like tiles on a roof. Therefore, such a recording cannot be called a full-fledged three-dimensional styling. Sequential recording writes data sequentially to tracks that are stacked on top of each other.

In this case, the disk layout does not undergo significant changes and standard magnetic heads can be used. All this helps to keep the price of hard drives low.

Disadvantage of SMR

The bottleneck of SMR technology is the rewrite procedure itself. When rewriting one of the tracks, you will have to rewrite all neighboring ones with which it overlaps. And this, in turn, significantly increases the recording time.

The lower the track is located in the tiled series, the more tracks will have to be overwritten. It turns out that no matter how well the logical volume is defragmented, the write speed can sag significantly.

In order to reduce the number of additionally rewritable tracks, they are grouped into structures that are called tapes. Overlapping tracks occurs only within the same tape. Therefore, if suddenly the data is located on the lowest “tile”, then the number of rewritable tracks will be predictable and the minimum rewrite speed will be determined in advance.

Partitioning into tapes is performed individually for each specific disk model, depending on the purpose of the device. Longer tapes are used on devices where data is not expected to be overwritten infrequently. They will be used either as archival drives or as devices for reading data.

Reducing the length of the tape will lead to a loss in the increase in the size of the drive, but will increase the speed of access to data located on the HDD.

Practical application in the budget segment

In the surviving technology of recording data on magnetic HDDs, manufacturers are still trying to somehow increase their capacity. Nevertheless, it is too early to write off mechanical HDDs: the cost of one gigabyte of memory on them is significantly lower than that of solid-state drives.

In the fight for storage space and price reduction, in 2014 Seagate released a relatively inexpensive and capacious drive model: ST8000AS002.

The disk has a size of 8 TB, which can be useful as a home media library. At the same time, the price for it is half that of other models of the same size (10 times lower than an SSD of the same volume).

This HDD is designed specifically for data storage. If it is initially written to, then data will be written at a speed that is not significantly inferior to other disks. And reading will be performed at quite acceptable speeds.

This model caused a controversial attitude among users, and so far one thing is clear: the technology has not passed by and is of interest to itself.

Summarizing

SMR technology still continues to improve: the principles laid down in it give room for building various variants of disks, optimized both in size and in speed. And when combined with techniques such as filling the disk chamber with helium, overall productivity can be increased.

As for the price, they are cheaper, about 1000-2000r compared to other HDDs. Depending on the store and where you live.

Seagate said late last year that hard drives will continue to evolve over the next 20 years, during which they will remain a relevant storage medium. The statement clearly hints that the company is developing a number of technologies that will increase the capacity and performance of hard drives for many years to come. Subsequently, Seagate announced the first HDDs with "tiled" magnetic recording, then introduced its first platform for helium-filled hard drives and a number of models with a capacity of 10 TB. In addition, it is no secret that in the coming years there will be hard drives based on two-dimensional magnetic recording (two-dimensional magnetic recording, TDMR) and thermal magnetic recording (heat-assisted magnetic recording, HAMR) technologies.

Some time ago, Mark Re, Seagate's senior vice president and chief technology officer, the person responsible for research and development, agreed to an interview with AnandTech to discuss Seagate's plans for the coming years. We publish a transcript of the conversation.

⇡#Evolution continues, new challenges arise

Although solid state data drives are evolving rapidly and increasing performance every year, in terms of cost per gigabyte, they in foreseeable future cannot compete with hard drives. Given this economic model, HDDs simply must continue to evolve, increasing storage capacity and performance.

Challenges for HDD Manufacturers Today

The evolution of hard drives has always included the development of a number of parameters: materials (platters), mechanics (motors, motors for heads, internal structure, etc.), read and write heads, controllers and firmware. The key factors influencing the growth of performance and capacity of hard drives have not changed over the years. We are talking about the size of the cell for storing the minimum amount of information, the width of the track (track), the speed of rotation of the spindle. The development of these components inevitably leads to the improvement of electronics, data transfer interfaces and other hardware inside and outside the HDD.

HDD recording density technologies to be used for the foreseeable future

The evolution of hard drives in the future will be based primarily on increasing the recording density of the plates, new heads, as well as on the computing capabilities of the controllers. Moreover, it is the latter parameter that will play a key role in the coming years. But what exactly is to be expected and when? Actually, this is what the conversation with Mr. Ree is about.

⇡#Seagate plans to expand the use of SMR plates

Today, the vast majority of hard drives are based on perpendicular magnetic recording (PMR) technology. The capabilities of PMR are quite sufficient for today's applications in terms of recording density and performance. A few years ago, hard drive manufacturers believed that PMR technology would not allow more than one terabit of data per square inch (Tbps) to be recorded. However, there has been some progress in the last couple of years, and it looks like the technology is still evolving. However, a more serious increase in recording density requires shingled magnetic recording (SMR) technology, which Seagate Technology Corporation began using several years ago for specialized drives.

Tiled Magnetic Recording: Basic Principles

SMR technology can increase recording density to above 1 Tbps, but brings with it a number of problems that need to be addressed. So, hard disks that use tiled recording record new tracks that “overlap” part of the previously recorded magnetic tracks. "Overlapping" tracks slow down recording because the SMR disk architecture requires new data to be written and then adjacent tracks to be overwritten. To minimize overwriting, the SMR-based hard disk controller arranges "tiled" tracks in groups (so-called bands). This optimizes the number of tracks that need to be rewritten after write operations and thus can help ensure predictable performance of SMR drives in typical scenarios.

Grouping tracks into tapes is not the only way to hide the features of tiled magnetic recording. In fact, every SMR platter-based hard drive has zones that use PMR recording technology with a relatively high write speed. These zones are used for fast data recording and a number of other operations when needed. Subsequently, information from PMR zones is automatically moved to SMR zones without any action on the part of the user or the operating system (which is akin to SSD garbage collection operations). Seagate does not disclose the actual configurations of its SMR lanes or PMR zone capacities, but notes that such configurations depend on the types of applications for which specific hard drive models are intended (i.e. similar characteristics differ for consumer HDDs and cold storage drives). data).

To further ensure optimal performance, SMR-based hard drives also use large DRAM and/or NAND flash based caches. For example, the 2TB Seagate Mobile 2.5” hard drive announced earlier this year features 128MB of DRAM and an unspecified amount of single-bit cell (SLC) NAND flash. The SLC NAND buffer has a fairly high write speed, which means that, in the case of working with small amounts of data, the above drive will boast a very high write speed. Since the amount of NAND is not very large (less than one gigabyte in the case of the mentioned model), this will not help in any way in the case of writing large files. But for domestic needs, such an architecture should be quite useful.

One of the things Seagate is proud of is its iterative approach to optimizing the write performance of hard drives based on SMR platters. It can be argued that the declared performance figures for the Seagate Archive 8 TB and Seagate Mobile 2 TB are not impressive. However, one cannot fail to note the implementation of three-level caching in Seagate Mobile 2 TB, which includes PMR zones, NAND and DRAM, which demonstrates the rather high complexity of such HDDs. Clearly, the architecture of SMR-based client drives requires high performance controllers to manage buffers, transfer data from PMR zones to SMR zones, and perform other operations, all to ensure expected performance in a variety of workloads. We've already seen something similar with TLC NAND based SSDs, which use pseudo-SLC based buffers to enable fast writes. Depending on Seagate's plans, the architecture of SMR-based hard drives (meaning the so-called device-managed drives that fully manage their work without the participation of third-party applications, operating system and / or hardware) can be expanded to further increase performance.

For the foreseeable future, Seagate plans to expand the use of tiled magnetic platters. Seagate will soon introduce SMR-based hard drives for video surveillance applications (i.e., optimized for streaming recording from a large number of surveillance cameras - at least 32-64). Further "tiled" hard drives for client devices will follow. There is no certainty that SMR technology will be used for high performance applications. First, because PMR continues to evolve; secondly, due to the inevitable emergence of new technologies, which we will discuss below. However, we could very well see some kind of hybrid drives that would use both SMR and PMR to guarantee high performance. In any case, Seagate does not believe that SMR can be used exclusively for storing rarely used "cold" data.

⇡#Helium will remain exclusive to high-capacity HDDs

Typically, companies tend to introduce new technologies to high-end products (enthusiast or enterprise) first, and then apply them to all other product families, such as client PCs or specialized devices. Over time, something that was once an exclusive feature of expensive, "extreme" devices becomes an integral part of mass products. To some extent, we will see this in the case of hard drives filled with helium. However, not all so simple.

The main advantages of helium-filled HDDs. HGST slide

The density of helium is seven times lower than the density of air, which reduces the friction force acting on the magnetic plates inside the hard drives, and also reduces the force of gas flows affecting the positioning accuracy of heads and platters. Filling hard drives with helium makes it possible to install up to seven platters in them, reduce the power of the spindle motors, increase the accuracy of head positioning, and reduce noise and heat generation. All of these benefits are essential for today's data centers. HGST introduced the world's first commercial helium hard drive back in 2013, and Seagate began selling its 10TB helium drives this spring.

In November last year, Seagate said that it began experimenting with helium in the early 2000s and had 12 years of experience with it at the end of last year. Mark Rea reaffirmed that Seagate is really familiar with helium and that its sealed HDD platform is very reliable. However, the commercialization of the latter is at an early stage. So, at the moment, Seagate does not even have a marketing name for it (Western Digital calls its platforms HelioSeal).

While filling hard drives with helium helps to more accurately position heads (important as track widths and cell sizes shrink), Seagate says the company uses purely mechanical solutions to reduce the force of gas flows inside hard drives and plans to further improve them in the future. Thus, the use of helium is not something mandatory for the next generations of hard drives, which will use HAMR, TDMR and other technologies to increase the write density and read speed.

Seagate believes that maximizing hard drive capacity (which automatically increases server rack and data center capacity) and minimizing power consumption are essential for data centers in the first place (which is exactly why seven platters and low power motors make sense in this market segment). However, since the reduction of gas flow forces can be achieved in various ways, the use of helium may not be necessary for those HDD platforms that are not designed to create products of maximum capacity.

While Seagate may not appear to be very enthusiastic about sealed HDDs, it should be remembered that large corporations always develop a whole range of technologies and platforms and then use them when the time comes. Thus, if Seagate is not going to use helium for relatively inexpensive client HDDs now, this does not mean that the company will not be able to introduce such devices in the future. More recently, the company announced its Data Guardians series of hard drives, whose flagship 10-terabyte models - BarraCuda, IronWolf and SkyHawk - are filled with helium. Of course, we are talking about reusing the server platform introduced earlier this year, but it is pretty clear that the company is quite ready to use helium outside of the data center drive segment.

Seagate's competitor, Western Digital Corporation, makes extensive use of HelioSeal technology for a wide variety of applications. So, in the spring of this year, the company introduced WD Red, WD Red Pro and WD Purple helium-filled hard drives for NAS and video surveillance systems. In addition to this, she announced an external hard drive My Book 8 TB, which is filled with helium, but has a spindle speed of only 5400 rpm. This suggests that HelioSeal technology is becoming less expensive.

It's worth noting that while Seagate isn't revealing a forward plan for its helium-filled HDDs right now, Mark Rea has made it clear that such a plan exists.

⇡#TDMR hard drives to hit the market in 2017

Two-dimensional magnetic recording (TDMR) is another technology that should help increase the recording density and performance of hard drives. Seagate believes that TDMR will help increase recording density by 5-10% (however, the essence of the technology is much more important than an immediate increase in recording density, which we will discuss in more detail below). Plans to use TDMR for commercial hard drives were announced last September, and Mark Rea confirmed that Seagate is on track to implement them. The first HDDs using 2D magnetic recording technology will hit the market as early as 2017.

Key Benefits of 2D Magnetic Recording

TDMR promises to enable hard drive manufacturers to increase the density of HDD platters by making tracks narrower and cells smaller. New technologies make it possible to minimize the size of the writer head (writer), however, reading in this case becomes a difficult task. With a significant increase in the density of tracks on the magnetic plate, the reader head (reader) turns out to be wider than the track and it becomes more difficult for it to “read” the data on each track due to interference from adjacent tracks (inter-track interference, ITI), which interferes with signal recognition. To combat the ITI effect, TDMR technology involves the use of an array of readers that will read data from one or more tracks at the same time (the method has been described in a number of scientific publications). Reading data from the same track by two readers allows the HDD controller to reduce the noise ratio (signal-to-noise ratio) from different tracks and read the data correctly. Of course, this will require high processing power from the controllers, as well as error control (we can probably see the LDPC method in the case of hard drive controllers). Of course, theoretically, a single reader head can make more than one "pass" over the necessary sectors during read operations, as a result of which the controller will receive the required amount of data. However, such an algorithm will inevitably increase delays, require a large amount of on-board memory and increased computing power. In addition, it can cause a decrease in overall performance.

The increased number of readers on the HDD head will become more important in times of heat-assisted magnetic recording (HAMR): heating the wafer surface will reduce the track width and cell size, which will lead to an increase in recording density and ITI. Thus, directly HAMR technology will increase the recording density, and TDMR will provide an opportunity to solve problems with ITI interference.

In addition, Mark Rea said that with proper programming, hard drives with multiple readers on the same head will get increased performance. We are talking about the fact that readers will be able to read data from adjacent tracks at the same time, which will increase the reading speed for large amounts of data. This obviously won't make the new hard drives as fast as SSDs, but it will help Seagate customers increase the performance of their storage systems. At the moment, the company is not talking about its plans to use multiple readers to increase performance in commercial HDDs, since such devices will not appear tomorrow, if at all - but it is considering such a possibility.

Mark Rea confirmed that TDMR allows HDD manufacturers to increase the recording density by about 10%, which is quite significant compared to conventional PMR-type platters. However, additional storage capacity will not be "free", at least in terms of processing power of hard drive controllers. An array of readers at the heads increases the bandwidth requirements of the controller, as well as the amount of information that needs to be processed. As a result, the entire TDMR platform becomes very expensive: it uses a large number of readers, new platters, new motors and new controllers. That is why Seagate plans to use such platforms primarily for server drives sometime in early 2017. Seagate has not confirmed whether such hard drives will use both TDMR and helium, but Mark Rea said that virtually all technologies can be used together within a single HDD platform to create relevant solutions for a variety of applications. However, remember that we are talking about plans, and they often change.

⇡#New generation of 10K and 15K hard drives on the way

Today, high speed hard drives are not the fastest storage devices. However, they are used for servers of continuous operation, the replacement of which is quite rare due to the use of specialized software and the extreme complexity of the process. Increasing performance for such systems is an important component. That is why Seagate Corporation is preparing a new generation of hard drives with a spindle speed of 10 and 15 thousand revolutions per minute (10K and 15K revolutions per minute, RPM).

Dell PowerEdge Server. Dell Photos

Many running servers of continuous action (mission critical, MC) will continue to rely on hard drives with high spindle speeds. Such HDDs use the SAS (Serial Attached SCSI) interface with all its advantages for such machines - these drives do not amaze with performance, but are used everywhere. They won't be taken out of service anytime soon, which is a good sign for HDD manufacturers. However, the market (total available market, TAM) for ultra-fast 10K and 15K hard drives has shrunk in recent years precisely because of SSDs. But that doesn't mean that fast hard drives aren't being developed anymore. all over the place - in fact, Seagate is preparing another generation of such devices.

10K RPM Seagate Enterprise Performance Hard Drive

The new generation of Seagate Enterprise Performance 10K hard drives will have not only a 10,000 rpm spindle speed, but multiple readers per head. Both heads will read the same track due to the very high recording density (remember that 2.5-inch hard drives can use smaller platters, so we can deal with something above 1 Tbps2), thus guaranteeing expected performance of a 10K drive by eliminating ITI. Considering that in the case of 10K hard drives we are dealing with critical data storage devices, we should expect that they will be based on already familiar PMR or SMR platters. However, multi-reader heads may eventually appear in HAMR-based HDDs.

Seagate Cheetah 15k rpm hard drive

As for hard drives with a spindle speed of 15 thousand revolutions per minute, Seagate behaves somewhat more modestly and secretively. The company confirms that it is working on another generation of 15K HDDs, but is reluctant to reveal details. It should be remembered that there are many companies using a large number of 15K SAS hard drives in the data center, and therefore another generation of such drives would come in handy. At the same time, SNIA has an ambitious long-term plan to develop the SAS interface and increase data transfer rates to 24 Gb / s by 2020 and beyond. Therefore, it is important for Seagate to offer both SSDs and HDDs for this market segment. It should be remembered that sales of 15K HDDs have been declining in recent years and the new generation of such devices may be the last - which is why it should offer a certain set of features and technologies that Seagate does not want to discuss. Perhaps not so much because of competition, but because the company is working with its customers to include features in the new HDD that are really needed.

However, as new data center platforms emerge, the need for 10K/15K hard drives will inevitably fall. For example, all new SSDs manufactured by Intel for data centers and continuous servers are designed for the PCIe interface and the NVMe protocol. As one of the largest manufacturers of high-end NVMe SSDs (Seagate Nytro family of accelerators), Seagate will follow the market trends. What's more, the company's recent SSD demonstrations (such as the 60GB SSD and the 10GB/s SSD) confirm Seagate's understanding and development of SSDs.

⇡#HAMR: over 2 Tbps and beyond

As mentioned above, SMR and TDMR technologies are capable of increasing the recording density on hard disk platters by 10-20% compared to today's recording density. Seagate has done a lot to make SMR technology the most viable for a wide variety of hard drive types, and in the future we will see the fruits of TDMR. However, given the physical limitations of SMR and TDMR, as well as the relatively high cost of helium-filled HDDs, a new magnetic recording technology is required to significantly increase storage capacity. Finally (we know you've been waiting for this particular section :)), it's time to talk about thermomagnetic recording technology - HAMR.

Key Benefits of HAMR Technology

Seagate says prototypes of its HAMR-based hard drives use heads that heat the platter locally to 450°C using a 20mW 810nm laser. Currently, HAMR-based hard drives have a recording density of approximately 2 Tbps, which is significantly higher than today's hard drives based on PMR or SMR platters. Potentially, this means that Seagate can double the capacity of hard drives just by using HAMR technology. In fact, not everything is so simple.

A device that transmits thermal radiation to heat a data carrier is called a near field optical transducer (NFT). When exposed to a laser, the NFT transfers thermal energy to the plate, thereby expanding the cells and enabling recording. Hard disk manufacturers use gold as the primary material for NFTs due to its excellent optical properties. On the other hand, gold has a relatively low mechanical strength, and such NFTs can deform when exposed to elevated temperatures for a long time. In turn, deformation can lead to a decrease in the ability to transfer thermal energy to the media, which, in fact, means a hard drive failure. This is why Seagate and other hard drive manufacturers have been researching and patenting various materials (gold-based alloys, to be precise) for NFTs over the years. Of course, Seagate does not disclose the alloy that is being used in prototype HAMR HDDs right now.

However, Mark Rea emphasizes that when the company ships the first HAMR-based hard drives to its partners for evaluation in 2017 and then to commercial systems in 2018, they will be built to last for a long time, just as like today's HDDs. Seagate does not disclose specific data on the capabilities of HAMR-based hard drives, but claims that they will be able to overwrite data several times a day for five years, which indicates a fairly high reliability. Eventually, client PC drives will also use HAMR, but such devices will appear relatively soon.

In addition to a solid NFT, HAMR-based hard drives will need new heads (with a heater, writer, and possibly multiple readers to combat the ITI effect), which means a lot of hardware work on multiple fronts. In addition, more powerful controllers and firmware will be needed. As expected, HAMR will allow to increase not only the capacity, but also the performance of hard drives. However, for this Seagate will have to develop a rather complex platform, which will include new media materials, new heads, advanced controllers and a number of other things.

It should be noted that HAMR is a challenge for the entire industry, not just Seagate. As a result, once the industry figured out how to make HAMR technology hard drives as reliable as traditional hard drives, the technology would begin to be mass-produced.

⇡#Summing up

The evolution of consumer hard drives in recent years hasn't been particularly exciting, but things have begun to change this year. The use of SMR will help increase hard drive capacity in the coming quarters, and then TDMR will open new doors for the coming years. There is one thing that should be clear at this point: the evolution of hard drives in the future will be different from their evolution in the past. The reason for this was the segmentation of the HDD market and the need for specialization of models.

Seagate hard drive

For example, hard drives for archiving, nearline applications, NAS and DAS must have increased capacity. However, performance hardly matters for archive hard drives or DAS. At the same time, nearline and NAS should offer both capacity and relatively high performance due to the fact that they can be used by a large number of clients at the same time. The most logical way to ensure maximum capacity and performance today is to use a helium platform with a 7200 rpm motor. As can be seen from the latest announcements by Seagate and Western Digital, this approach is used for 10 TB hard drives (in the case of Seagate, these are Enterprise Capacity, Barracuda Pro, SkyHawk and IronWolf devices), when top models are built on a special platform. If we are talking only about the maximum capacity for a PC, then the Seagate SMR client platform is the best suited for such drives.

The situation is unlikely to change in the coming years, since the development of technologies for the next generations of hard drives requires significant investments against the backdrop of declining demand for HDDs. As a result, a number of technologies or combinations of technologies will not be used to build all types of hard drives (we will not see helium-filled HDDs in the low price segment). Some things will remain mostly in the data center, specialized and expensive drives (like helium), others will be strictly directed to client computers (hybrid drives).

What's more, Seagate and its competitors understand that HDDs can't compete with SSDs in terms of performance, especially in random read/write situations. Thus, although hard drives will get additional speed and capacity in the coming years, we should not expect that their performance will henceforth be the main point of concern for manufacturers. Recording density and power consumption are becoming new factors to be taken care of by Seagate, Toshiba and Western Digital.

Seagate's forward plan includes SMR, TDMR, HAMR, and various other platter recording methods. The company is developing a set of technologies that should increase the capacity, performance, reliability and endurance of future hard drives using the aforementioned recording methods. While Seagate is confident that its devices will be in demand, there are things that are difficult to predict: for example, we are not sure how the client storage market will develop. Be that as it may, time will tell.

Today, the growth in data per person is growing exponentially, and companies offering storage solutions for this data are striving to do everything possible to increase the available capacity of their devices. Seagate's Shingled Magnetic Recording (SMR) tile magnetic recording technology improves recording density, increasing disk capacity by 25%. This is possible by increasing the number of tracks on each plate and reducing the distance between them. Tracks are placed on top of each other (like tiles on a roof), which allows you to record more data without increasing the area of \u200b\u200bthe plate. When new data is written, tracks overlap, or are "truncated". Due to the fact that the read element on the disk head is smaller than the write element, it can read data even from a truncated track without violating their integrity and reliability.

However, the following problem is associated with SMR technology: in order to overwrite or update information, it is necessary to overwrite not only the required fragment, but also the data on the last tracks. Because the recorder is wider, it captures data on adjacent tracks, so you need to overwrite those as well. Thus, when changing the data on the lower track, you need to correct the data on the nearest overlay track, then on the next one, and so on, until the entire plate is rewritten.

For this reason, the tracks on an SMR disc are grouped into small groups called tapes. Superimposed on each other, respectively, only the tracks within the same tape. Thanks to this grouping, if some data is updated, not the entire plate will have to be rewritten, but only a limited number of tracks, which greatly simplifies and speeds up the process. For each type of disk, its own tape architecture is developed, taking into account the scope of its application. Each Seagate product line is designed for a specific application and environment, and SMR technology delivers the best results when used correctly.

Seagate SMR is the technology to meet the ever-increasing demand for additional capacity. Today, it is being actively improved and, in combination with other innovative methods, can be used to increase the recording density on next generation hard drives.

But first of all, it is necessary to understand some of the nuances of its application.

There are three types of devices that support tiled recording:

Autonomous (Drive Managed)

Working with these devices does not require any changes to the host software. All write/read logic is organized by the device itself. Does that mean we can just install them and relax? No.

Drives that implement Drive Managed write technology usually have a large amount of write-back cache (from 128MB per disk). In this case, sequential requests are processed in the write-around mode. The main difficulties faced by developers of devices and storage systems based on this recording technology are as follows:

1. The cache size is limited and as it fills up, we can get unpredictable device performance.
2. Significant latency levels sometimes occur when flushing the cache heavily.
3. Determination of sequences is not always a trivial task, and in complex cases we can expect performance degradation.

The main advantage of this approach is the complete backward compatibility of devices with existing operating systems and applications. With a good understanding of your task, you can buy Drive Managed devices now and benefit from the technology. Further in the article, you will see the results of testing such devices and be able to decide how they suit you.

Host Managed

These devices use a set of extensions to ATA and SCSI to interact with disks. This is a different type of device (14h) that requires major changes to the entire Storage Stack and is incompatible with classic technologies, that is, without special adaptation of applications and operating systems, you will not be able to use these drives. The host must write to devices strictly sequentially. At the same time, device performance is 100% predictable. But the correct operation of higher-level software is required in order for the performance of the storage subsystem to be truly predictable.

Host Aware

These are hybrid solutions that combine the benefits of Device Managed and Host Managed technologies. By purchasing such drives, we get backward compatibility support with the ability to use special ATA and SCSI extensions for optimal work with SMR devices. That is, we can both simply write to devices, as we did before, and do it in the most optimal way.

In order to provide work with Host Managed and Host Aware devices, a couple of new standards are being developed: ZBC and ZAC, which are included in T10 / T13. ZBC is an extension of SCSI and is ratified by T10. Standards are being developed for SMR drives, but may be applied to other devices in the future.

ZBC/ZAC define a logical device model where the main element is a zone, which is mapped as an LBA range.

The standards define three types of logical zones into which devices are divided:

1. Conventional zone - a zone with which we can work in the traditional way, like with ordinary hard drives. That is, we can write sequentially and randomly.

2. Two types of Write Pointer Zone:

2.1. Sequential write preferred - the main zone type for Host Aware devices, sequential write is preferred. Random writes to devices are handled like Device Managed devices and can cause performance degradation.

2.2. Sequential write only - the main zone type for Host Manged devices, only sequential write is possible. Random writes are not allowed, and attempts to do so will return an error.

Each zone has its own Write Pointer and its own status. For all devices that support the HM write type, the first LBA of the next write command must match the position of the Write Pointer. For HA devices, Write Pointer is informational and serves to optimize disk handling.

In addition to the new logical structure, new commands appear in the standards:

REPORT_ZONES is the main method through which you can get information about the existing zones on the device and their status. In response to this command, the disk reports existing zones, their types (Conventional, Sequential Write Required, Sequential Write Preferred), zone status, size, information about the location of the Write Pointer.

RESET_WRITE_POINTER is the successor to the TRIM command for ZBC devices. When it is called, the zone is erased and the Write Pointer is moved to the beginning of the zone.

Three optional commands are used to manage the zone status:

OPEN_ZONE
CLOSE_ZONE
FINISH_ZONE

New information has been added to the VPD pages, including the maximum number of open zones for better performance and the maximum number of zones available for random writes with better performance.

Storage manufacturers need to take care of supporting HA / HM devices by making changes at all levels of the stack: libraries, schedulers, RAID engine, logical volumes, file systems.

In addition, you need to provide two types of interfaces for applications to work: a traditional interface, organizing an array as a device managed device, and a virtual volume implementation as a HOST AWARE device. This is necessary as applications are expected to work directly with HM/HA devices.

In general, the algorithm for working with HA devices is as follows:

1. Define device configuration using REPORT_ZONES
2. Define areas for random recording
2.1. Quantity is limited by the capabilities of the device
2.2. In these zones, there is no need to track the position of the Write Pointer
3. Use the rest of the zones for sequential writing and using the position information of the Write-Pointer and doing only sequential writing
4. Control the number of open zones
5. Use Garbage Collection to Deallocate the Zone Pool

Some write techniques can be applied from existing all-flash storage systems, for which the problems of prostatic sequential writing and garbage collection were solved.

RAIDIX has tested Seagate SMR drives in its lab and provides some recommendations for their use. These drives differ in that they are Device Managed and do not require any major changes to the application.

During testing, an attempt was made to test the performance expectations of such drives and understand what we can use them for.

The tests involved two Seagate Archive HDDs with a capacity of 8000GB.
Testing was performed on the operating system Debian version 8.1
CPU Intel i7 c 2.67 MHz
16 GB RAM
The drives have a SATA 3 interface, we turned the controller into AHCI mode.

To begin with, we provide information about the devices by running an Inquiry query.

To do this, we used the sg3-utils set of utilities.

sg_inq /dev/sdb
standard INQUIRY:
PQual=0 Device_type=0 RMB=0 version=0x05
NormACA=0 HiSUP=0 Resp_data_format=2
SCCS=0 ACC=0 TPGS=0 3PC=0 Protect=0 BQue=0
EncServ=0 MultiP=0 Addr16=0
WBus16=0 Sync=0 Linked=0 CmdQue=0
length=96 (0x60) Peripheral device type: disk
Vendor identification: ATA
Product identification: ST8000AS0002-1NA
Product revision level: AR13
Unit serial number: Z84011LQ

On page 83 is the VPD.

sg_inq /dev/sdb -p 0x83
VPD INQUIRY: Device Identification page
Designation descriptor number 1, descriptor length: 24
designator_type: vendor specific , code_set: ASCII

vendor specific: Z84011LQ
Designation descriptor number 2, descriptor length: 72
designator_type: T10 vendor identification, code_set: ASCII
associated with the addressed logical unit
vendor id: ATA
vendor specific: ST8000AS0002-1NA17Z Z84011LQ

We didn't see anything special. Attempts to read zone information have failed.

RAIDIX makes software for storage systems working in various industries, and we tried not to use specialized or paid benchmarks.

We start by checking the streaming performance of disks on internal and external tracks. The test results will give the maximum expected performance of the device and are primarily consistent with tasks such as data archiving.

We did not touch the settings of the block subsystem. We perform testing by writing data to disks in blocks of 1 megabyte. For this we use the fio v.2.1.11 benchmark.

Jobs differ from each other only by their offset from the beginning of the device and are launched one after the other. libaio is chosen as the I/O library.

The results look good:

Performance on external and internal tracks differs by almost 2 times.
We see intermittent performance dips. They are not critical for archiving, but can be a problem for other tasks. With the correct operation of the write-back cache of the storage system, we assume that we will not observe such a situation. We ran a similar experience, creating a RAID 0 array of both drives, allocating 2GB of RAM cache to each drive, and saw no performance dips.

When reading the failures are not visible. And subsequent tests will show that SMR disks do not differ in performance from ordinary disks in read operations.

Now we will conduct more interesting tests. Let's run 10 threads with different offsets at the same time. We do this in order to check the correctness of buffering and see how the disks will work on CCTV, Video Ingest and similar tasks.
The graphs show the total productivity for all jobs:

The disk handled the load well!

Performance stays at 90 MB/s, evenly distributed across threads, and there are no major dips. The schedule for reading is absolutely similar, only raised by 20 MB. For storing and distributing video content, exchanging large files, the performance is suitable and practically does not differ from the performance of conventional disks.

As expected, the disks performed well in streaming reads and writes, and multithreading was a pleasant surprise for us.

Let's move on to "random" reading and writing. Let's see how disks behave in classic enterprise tasks: storing DBMS files, virtualization, etc. In addition, frequent work with metadata and, for example, enabled deduplication on an array fall into "random" operations.

We are testing in blocks of 16 kilobytes and are still correct fio.
In the test, we set up several jobs with different queue depths, but we will not give the full results. Only the beginning of the test is indicative.

The first 70.5 seconds we see an unrealistic 2500 IOps for a hard drive. This causes frequent failures. Apparently, at this moment, a buffer is written and periodically reset. Then there is a sharp drop to 3 IOps, which lasts until the end of the test.

If you wait a few minutes, then after the cache is reset, the situation will repeat.

It can be expected that with a small number of random operations, the disk will behave well. But if we expect an intensive load on the device, it is better to refrain from using SMR disks. RAIDIX recommends moving all metadata work to external devices whenever possible.

What about random reading?
In this test, we limited the response time to 50ms. Our devices are doing well.

The reading is in the range of 144-165 IOPs. The numbers themselves are not bad, but the spread of 20 IOPs is a little scary. Focus on the bottom line. The result is not bad, at the level of classic discs.

Let's change our approach a bit. Let's take another look at working with a large number of files.
The frametest utility from SGI will help us with this. This benchmark is designed to test the performance of your storage system when editing uncompressed video. Each frame is a separate file.

We have created an xfs filesystem and mounted it with the following options:
-o noatime,nodiratime,logbufs=8,logbsize=256k,largeio,inode64,swalloc,allocsize=131072k,nobarrier

Run frametest with the following parameters:

./frametest -w hd -n 2000 /test1/

The benchmark creates 2000 8MB files.

The start of the test goes well:

Average details:

Last 1s: 0.028ms 79.40ms 79.43ms 100.37MB/s 12.6fps
5s: 0.156ms 83.37ms 83.53ms 95.44MB/s 12.0fps

But after recording 1500 frames, the situation worsens significantly:

Average details:
Open I/O Frame Data Rate Frame Rate
Last 1s: 0.035ms 121.88ms 121.92ms 65.39MB/s 8.2fps
5s: 0.036ms 120.78ms 120.83ms 65.98MB/s 8.3fps

Average details:
Open I/O Frame Data Rate Frame Rate
Last 1s: 0.036ms 438.90ms 438.94ms 18.16MB/s 2.3fps
5s: 0.035ms 393.50ms 393.55ms 20.26MB/s 2.5fps

Let's do a reading test:

./frametest -r hd -n 2000 /test1/

Throughout the test, the performance is excellent:

Average details:
Last 1s: 0.004ms 41.09ms 41.10ms 193.98MB/s 24.3fps
5s: 0.004ms 41.09ms 41.10ms 193.98MB/s 24.3fps

Currently, work is underway on specialized file systems for SMR disks.
Seagate is developing an ext4-based SMR_FS-EXT4. It is possible to find several log-structured file systems designed specifically for Device Managed SMR drives, but none of them can be called a mature, recommended product for implementation. Seagate is also developing a Host Aware version of the SMR drive, which should be completed before the end of the year.

What conclusions can we draw from the results of performance measurements?
Device Managed devices can be safely used for tasks that do not differ in intensive recording. They cope very well with the tasks of single-threaded and multi-threaded recording. They are great for reading data. Periodic "random" disk requests for metadata updates are consumed by a large cache.

For solving problems characterized by intensive “random” recording or updating a large number of files, such devices are not very suitable, at least without the use of additional technical means.

The MTBF parameter of the tested drives is 800,000 hours, which is 1.5 times lower than that of, for example, NAS drives. The large volume of disks significantly increases the recovery time and makes regular media-scans almost impossible. We recommend that when designing storage with such drives, rely on RAID with a parity greater than 2 and/or approaches that reduce rebuild time (eg, Parity Declustering).

The Shingled Magnetic Recording (SMR) technology developed by Seagate specialists will soon allow increasing the data density on hard disk platters by 25% due to a fundamentally new track layout. Next year, mass production of 3.5-inch hard drives with a capacity of 5 TB will be launched, and by 2020 the maximum volume of such drives will reach 20 TB.

information explosion

According to experts, the world's population of about 7 billion people currently generates a total of 2.7 zettabytes of data annually. And you don't need to be an information technology specialist to understand that this figure will only increase with each subsequent year. One of the factors contributing to this is the increase in the bandwidth of the channels used to connect to the Internet both via fixed lines and through public wireless access areas and cellular networks. Year by year, the amount of data (and, above all, media files) being uploaded to cloud storage, as well as stored on the hard drives of home PCs and NAS drives, is increasing. And this is quite natural. First, the resolution of household photo and video cameras increases, and, consequently, the volume of stored images and video recordings with the same number of shots and video timing. Secondly, due to the increase in the bandwidth of Internet access channels, it has become possible to stream media content of much higher quality. Naturally, high-definition video (especially in stereoscopic format) requires much more storage space than standard-definition files.

A serious factor that creates an additional load on data storage systems is the rapid growth of the fleet of mobile devices - primarily smartphones and tablet PCs. Since such gadgets, as a rule, are equipped with a relatively small amount of internal memory, their owners often need to use external drives to store both self-generated and externally downloaded media content.

According to John Rydning, vice president of market research for hard drives at market research firm IDC, the hard drive industry is currently experiencing a period of significant growth. The total capacity of the supplied drives is measured in petabytes, and the annual increase in this indicator is about 30%. However, at the same time, developers manage to increase the specific density of magnetic recording by less than 20% per year.

Thus, despite the constant improvement of the technologies used in hard drives, the manufacturers of these components do not keep up with the rapidly growing market needs. However, one can hardly blame the developers for this, who are already tirelessly looking for more and more new ways to increase the magnetic recording density.

For example, in 2007 Seagate was the first company to introduce Perpendicular Magnetic Recording (PMR) technology in commercial hard drives. Due to the orientation of the magnetic domains not parallel to the disk plane, but perpendicular to it, it was possible to reduce the size of the track and thereby increase the capacity of one plate up to 250 GB.

Five years later, thanks to the systematic development of this technology, it was possible to increase the specific density of magnetic recording by four times and fit 1 TB of data on one plate. This achievement has led to the mass production of 3.5-inch hard drives with a capacity of 4 TB. However, under the current conditions, this is no longer enough.

One way to bridge the widening gap between user demand and hard drive performance is to introduce Shingled Magnetic Recording (SMR) technology developed by Seagate. Let's see what the essence of this solution is.

Shingles principle

Most readers probably know that data on the surface of hard disk platters is recorded on the so-called tracks, which can be simplified as a set of concentric circles (Fig. 1). The smaller the width of the tracks and the intervals between them, the higher the specific recording density, and hence the capacity of the drive with the same form factor and number of platters.

Rice. 1. Track layout
on the surface of the magnetic plate

With the traditional method of magnetic recording, the minimum track width is determined by the physical dimensions of the recording element of the hard disk head (Fig. 2). To date, the limit of miniaturization of magnetic head elements has already been reached, and a further reduction in their size using existing technologies is impossible.

Rice. 2. With the traditional layout of the tracks, their minimum width
limited by the size of the recording element of the magnetic head of the drive

SMR technology allows to bypass this limitation and increase the specific recording density due to a denser arrangement of tracks, which are partially superimposed one on top of the other like elements of a tiled roof (Fig. 3). As new data is being written, tracks with previously saved data are cut off as if. Since the width of the reading element of the magnetic head is less than the width of the recording element, all the data on the plate can still be read from the trimmed tracks without compromising the integrity and safety of this information.

Rice. 3. When using SMR technology, the tracks are arranged more closely,
overlapping one another

While everything is simple and clear. However, if you need to write new data over existing ones, a problem arises. After all, in this case, you will have to overwrite not only this fragment directly, but also data blocks on the following tracks. Since the recording element of the magnetic head is wider than the reading element, the overwriting process will destroy data previously stored on adjacent areas of adjacent tracks (Fig. 4). Thus, to ensure the integrity of previously recorded information, these blocks must first be buffered and then written back to the appropriate track. Moreover, this operation will have to be repeated sequentially for all subsequent tracks - until the border of the working area of the magnetic plate is reached.

Rice. 4. In the process of overwriting data on one
of the tracks, a section of the adjacent track will be affected

With this feature in mind, the tracks in hard drives with SMR technology are divided into small groups - the so-called packages (Fig. 5). This approach provides more flexible control over the process of adding and overwriting data, and most importantly, allows you to reduce the number of additional overwriting cycles and thereby increase the performance of the drive. Even if the packet is already full, then when replacing a data block in it, it will be necessary to rewrite sections of only a limited number of tracks (up to the border of this packet).

Rice. 5. Layout of the tracks in the package

The structure of the packages on the drive may be different depending on the scope of a particular model. Thus, for each family of hard drives, you can create a unique package structure, optimized for the specific use of these drives.

It is important to note that the introduction of SMR technology does not require significant changes in the design of magnetic heads and restructuring of the production process of these components. This will keep the cost of new drives at the same level, and due to the higher capacity, achieve even more attractive indicators for the unit cost of data storage.

Conclusion

So, SMR technology is a very effective solution that allows you to meet the growing demand for increasing the maximum capacity of hard drives in a short time and at minimal cost. At the first stage of implementation of SMR technology, it will increase the data recording density by 25% - from 1 to 1.25 TB per 3.5-inch plate. Thus, next year it will be possible to produce hard drives with a capacity of 5 TB.

It is important to note that in the case of the introduction of SMR technology, an increase in the capacity of drives is achieved without increasing the number of magnetic heads and/or hard disk platters. Thus, new hard drives with a higher capacity will be as reliable as previously produced models of a similar form factor. In addition, as mentioned above, the introduction of SMR technology does not require significant changes to the design of the hard drive. This, in particular, allows the use of the same magnetic heads and plates that are installed in the current models.

Another advantage of SMR is the ability to combine this solution with various magnetic recording technologies. Currently, it is used in hard disks with perpendicular magnetic recording, but in the future it can be used in combination with other solutions that will allow to achieve even higher specific recording density.

Article based on materials from Seagate