June 8, 2016 at 12:01

How to choose hard drives for servers?



    There are many myths in the IT field. “You can unsubscribe from spam,” “Two antiviruses are better than one,” “Server hard drives should only be branded.” When replacing and expanding the railway fleet, many nuances and subtleties must be taken into account, and it was not without its prejudices. What are the hard drives for servers, how they differ, what you need to pay attention to, and whether they should be with the logo of the server manufacturer - read about this under the cat.

    If the disk is installed in the server, then it must meet strict requirements for:

    • Reliability . Unrecoverable data loss can result in multimillion-dollar losses and reputation losses.
    • Performance . Servers are a priori designed to handle numerous requests.
    • Response time . Users do not have to wait until the server disk wakes up and processes their requests.

    In other words, the hard disk in the server should be like a pioneer - it is always ready to process numerous requests with a minimum level of delay, providing a high level of data security. In heavily loaded servers, hard drives have been working intensively and non-stop for years.

    There are four main categories (we do not take into account SSD, SAS SSD, PCI-e SSD) hard drives:

    • SATA (ordinary, "household" SATA) - spindle speeds of 5400 and 7200 rpm.
    • SATA RAID Edition (SATA RE) - 7200 rpm spindle speed, support for RAID controller commands.
    • SAS Near Line (SAS NL) - spindle speed 7200 rpm.
    • SAS Enterprise - spindle speed of 10,000 or 15,000 rpm.

    First of all, you need to decide on the connection interface - SATA or SAS.

    SATA or SAS?




    The SATA interface is a development of the IDE, which was later renamed PATA . That is, this interface is initially focused on use in household computers, as well as in industrial systems with moderate performance and reliability requirements. At the same time, SAS is the successor to the classic "server" SCSI interface .

    Initially, the SAS interface had higher bandwidth than SATA. But progress does not stand still, and the third generation SATA III has a maximum throughput of 6 Gb / s, as well as the second generation SAS. However, servers with a third-generation SAS controller are already available on the market, with a throughput of up to 12 Gb / s.

    To connect SAS disks, the server must be equipped with an appropriate controller. At the same time, backward compatibility of interfaces is ensured: SATA disks can be connected to the SAS controller, but vice versa - it is impossible.

    SAS provides full-duplex data exchange: the hard drive processes one read and write command at a time, and the SATA drive either read or write. But this advantage will be noticeable only with a large number of drives, if you compare SAS NL and SATA RE.

    To summarize: SATA-drives are good for creating large storage, which does not require maximum performance. And if you need to squeeze out everything possible from the disk subsystem, then your choice is SAS.

    Say no to regular hard drives




    We’ll clarify right away - regular SATA is not intended for use in servers. There are several reasons for this:

    • Low vibration resistance.
    • High level of unrecoverable errors.
    • Lack of support for hardware RAID controller commands.

    Of course, the cost of conventional desktop SATA is significantly lower than that of server SATA, and nothing prevents them from being used for small tasks that do not require high-performance disk subsystems. If the safety and speed of access to data is in the first place, then we strongly recommend taking server-side railway drives.

    Resistance to vibrations


    To solve more or less demanding tasks, it makes no sense to put only one disk. To ensure the minimum level of reliability of data storage, you need at least two drives combined in RAID. But when 4 or more devices are assembled in the basket, the vibrations arising from their operation affect the stability of spindle rotation and the accuracy of head positioning. Therefore, server hard drives have a number of design differences from household ones:

    • Reinforced spindle shaft, more resistant to external influences.
    • Extra vibration control.
    • Technologies that significantly increase the accuracy of positioning and the flight height of the heads above the surface of the “pancakes”.
    • Rich self-diagnosis capabilities that allow you to notify on the imminent failure of the drive in time.

    Household drives are devoid of all this. With a sufficiently strong vibration level, the probability of read / write errors in conventional SATA is 50% higher than in SATA RE .

    Fatal Error Level


    The next difference between server hard drives and household hard drives is the level of unrecoverable errors. For ordinary SATA, it is about 10 -14 (1 bit for every 10 14 bits read = 12.5 terabytes). That is, with a sixfold rewriting of a two-terabyte disk, you almost certainly get one unrecoverable error. For household drives, this is not a problem. But if you rewrite the database every month, then in six months it may turn out to be a bit. Where:

    Вероятность возникновения невосстановимой ошибки =
    (N * (X / 12500 * 12500) / 12500) * 100%



    • N is the number of disks in the RAID array
    • X is the volume of one disk in gigabytes,
    • 12500 - the number of bits per 1 irreparable error, expressed in gigabytes.

    Suppose you created a RAID 5 array of 1 terabyte regular SATA. If you rebuild the array, you will receive an unrecoverable error with a probability of 40%. And if you are using a 600-gigabyte drives, the probability of epic fail at Rebild is 24%: I and SATA RE SAS NL nonrecoverable error rate of 10 -15 , that is an order of magnitude smaller than conventional SATA. Then in our example with RAID 5 we get: For 1-terabyte disks, the probability of error = 4%. For 600 GB drives, the probability of error = 2.4%. For SAS disks, the level of unrecoverable errors is even lower - 10 -16 : For 1-terabyte disks, the probability of error = 0.4%. For 600 gigabyte disks, the probability of error = 0.24%. note

    Вероятность возникновения невосстановимой ошибки =
    (5 * (1000 / 12500 * 12500) / 12500) * 100% = 40%.



    Вероятность возникновения невосстановимой ошибки =
    (5 * (600 / 12500 * 12500) / 12500) * 100% = 24%.











    : The probability of an error is proportional to the number of disks in the RAID array.



    How do SAS drives provide lower error rates? Magic .

    • The sector size in SATA disks is 512 bytes, in SAS disks - 520 bytes. An additional 8 bytes is used for pass-through parity.
    • Other reading algorithms.
    • Additional data recovery algorithms without the participation of the controller.

    Work in a RAID array


    Another important drawback of conventional SATA is the lack of error correction when working in a RAID array. Let's say you hoped for a backup, and for the sake of economy, built RAID from ordinary SATA. If an error occurs, the hard drive repeatedly tries to read the bad block. And while he does this, he does not respond to the signals of the RAID controller. He perceives this as a failure of the hard disk, excludes it from the array and tries to recover. In other words, when an error occurs, the entire disk drops out of the array.

    In the case of SATA RE, SAS NL and SAS, the situation will develop differently. Having found an error, the disk informs the controller about the presence of a bad block. The controller requests this block from other disks in the array and transfers it to the failed disk. At the same time, the device does not drop out of the array, and performance does not drop.

    The myth of brands


    Finally, the most important question: do I need to buy "native" drives?

    It is no secret that HP, IBM and DELL do not produce hard drives. They buy them from third-party manufacturers, after which they test, reflash and glue their logos.



    On the one hand, such disks have several advantages:

    • firmware (firmware) takes into account the features of the controllers of certain server models,
    • additional quality control and stress tests reduce the likelihood of acquiring specimens with hidden defects,
    • “branded” discs are provided with a vendor warranty and full support.

    But you have to pay for all the good - "native" hard drives sell about two to three times more expensive than the same models, but with the logos of the manufacturers - Seagate, Western Digital, Toshiba, HGST.

    As you know, such a difference in price is far from being justified by promises of increased reliability. Therefore, many have probably heard that non-native hard disks work in HP, IBM, and DELL servers unstable or too slowly. Someone even scares that the server will not start with the “left” disks.

    Where do these statements come from?

    The vast majority of servers use technologies to increase the performance of the disk subsystem. It is for this purpose that vendors reflash hard drives - to provide support for these technologies. If you put "non-native" discs, then you just can not use the branded know-how, nothing more.

    Also, before, vendors artificially forced to use "branded" drives, using locks at the controller level. In the end, the anger of the masses forced over time to abandon this vicious practice. Today, problems most often arise with relatively old server models. And this is solved by simply filling in the controller with fresh firmware. Although there are simply capricious models of controllers, for example, P410 in HP servers.

    As practice shows, non-native hard drives work without difficulty:

    • on HP servers - at least from the Gen6 generation,
    • on IBM servers - at least since the M2 generation,
    • in DELL servers - from at least 10 generation.

    Non-native hard disks are fully compatible with both server rails and internal monitoring systems. But the sleds need to be set only by relatives, and only for a certain generation of the server. You can easily find on the net the real names of the disc models that vendors sell under their own brands. For example, most HP SAS drives are made from the Seagate Savvio line of hard drives. At the same time, it is not at all necessary to look for complete analogs, you can choose suitable models from popular rulers:


    As for the reliability of a vendor, according to quite informative statistics from Backblaze , a cloud backup company , Hitachi drives are the most reliable. In second place is Western Digital, in third is Seagate.

    Check Warranty


    If you decide not to follow the vendor's plans and intend to buy non-native hard drives, then be sure to check with the seller first: who provides the guarantee? The fact is that many stores do not provide a guarantee for hard drives, referring to the manufacturer's warranty. But there is a subtle point: for example, for some Seagate models, the warranty period begins from the moment of production . Therefore, the situation is not ruled out that you will buy completely new discs that have already ended the manufacturer’s warranty.

    Sad, but true.

    In order not to experience this awkward moment, try to check the guarantee of specific items on the manufacturers' websites before buying:

    Seagate: http://support.seagate.com/customer/en-US/warranty_validation.jsp

    Western Digital: http://support.wdc.com/Warranty/warrantyStatus.aspx?lang=en

    Hitachi: https://www.hgst.com/portal/site/en/support/warranty

    Conclusion


    When choosing hard drives, you must first start from the tasks that the server will perform :

    • If you do not need high access speed and reliability of data storage, and the number of disks will not exceed four, then we recommend installing SATA RAID Edition disks. This is an option for low-cost entry-level servers serving a small number of users.
    • If the server will serve the database, or the number of disks in the array will be 5 or more, it is better to choose SAS NL. Most often, such disks are placed in servers operating in medium-sized companies: under accounting systems, CMS, corporate repositories, etc.
    • And if you need maximum performance and / or reliability of data storage, for example, when processing financial transactions, then your choice is SAS Enterprise drives. These are media for highly loaded servers serving a large number of users, as well as for systems that work with the most important data.

    But most importantly - do not believe myths. It’s not necessary to buy discs with the same logo as on your server. With a competent approach, you can significantly save on upgrading the disk subsystem, without losing a bit of reliability and speed.
    Tags:

    Also popular now: