Enterprise hard drives

Why enterprise hard drives

Servers are different from client computers (desktop computers, laptops, and tablets) in several ways:

  • Servers usually run continuously; client computers are often powered down at the end of the day.
  • Servers often have high workloads running at 80% or above; clients are typically running at around 10% to 20% workload.
  • Servers are business-critical (e.g. if the server fails your business halts); clients are relatively expendable (e.g. if your desktop computer fails you can log on using your laptop).

Because they’re so important to your business, servers generally cost much more than client computers. And a large chunk of the cost of a server is its storage subsystem. I was recently costing out a new server for our business and the vendor’s price tag looked like this:

  • Server system with everything except hard drives = about $1,500
  • Server system with everything including four enterprise-class hard drives = about $4,000

Note that the storage was almost two-thirds of the cost of the server!! Why so much? Because I needed hot-swappable HDDs and the vendor only supported using their own hot-swappable HDDs for this particular system. Had I been able to shop around on NewEgg or some other site for cheaper HDDs, I might have shaved $1,000 or more off the total price tag for my system. But vendor lock-in together with my business need prevented me from doing that on this particular occasion.

But apart from the vendor lock-in issue, why are enterprise-class HDDs generally so darn expensive? It all has to do with the differences between client computers and servers summarized above. For example:

  • Because servers need to be highly reliable, their HDDs need to have high reliability. That means enterprise-class HDDs must fail very infrequently, or in more specific terms, their Mean Time Between Failure (MTBF) must be much higher than desktop-class HDDs.
  • Because servers are often continuously running under heavy workloads, they need high-performance HDDs. Enterprise-class HDDs must therefore include mechanisms that enable data to be accessed, read, and written faster than your typical desktop-class HDD. This means that enterprise-class drives should have a larger cache, higher rotational speed, and faster interface than typical desktop drives.
  • Because the data stored on servers is usually critical to the success of your business, their HDDs must be highly resistant to corruption. On a client computer, one bad sector on the HDD might cause Windows to hang or crash, but it usually won’t result in data loss for your business because business data is usually stored on network file servers, not client computers. A bad sector on a server’s HDD however could result in an erroneous business transaction that could alienate customers and cost your company money. Enterprise-class HDDs must therefore include more resilient technologies for quickly identifying, repairing, and remapping bad sectors to prevent extended service interruptions for clients. Enterprise HDDs also include additional error-checking features to ensure greater data integrity than desktop HDDs typically provide.
  • Because servers run continuously and typically have multiple HDDs in a RAID configuration, they also need drives that are power-optimized so they don’t consume excessive amounts of AC power. This is especially true in datacenter environments where you have racks of blade servers. HDDs for such servers typically use the 2.5 inch format instead of the 3.5 inch format typically used in desktop computers because the smaller platter size generally consumes less AC power.

The bottom line is this: if your data is important to your business, then your servers should use enterprise-class drives. And since you shouldn’t be storing any business-critical data on client computers, your client computers should NOT be using enterprise-class drives.

Types of enterprise-class drives

There are several different ways of classifying enterprise-class drives. For example, you could classify them according to technology. Thus, you have hard disk (i.e. spinning platter) drives (HDDs) and solid state drives (SDDs) that are designed for use in server systems. And if we focus only on HDDs, then you have Serial Attached SCSI (SAS) drives and Serial ATA (SATA) drives designed for server systems.

But there is another key way of differentiating between different enterprise-class drives, and that has to do with the tradeoff between performance and capacity. Specifically, when you’re buying an enterprise-class drive you usually have to decide between one of the following two types of drives:

  • Performance-optimized drives – These drives are designed to seek and transfer data as fast as possible. For HDDs this typically means they must have low seek time and rotational latency, and should have a sequential throughput of 150 MB/sec or higher. An example of a performance-optimized HHD might be a 500 GB 10,000 RPM 2.5 inch SAS drive with a 6 Gb/s transfer rate.
  • Capacity-optimized drives – These drives are designed to hold as much data as possible. For HDDs this typically means they have moderate seek time and rotational latency, and should have a sequential throughput of at least 100 MB/sec. An example of a capacity-optimized HDD might be a 4 TB 7200 RPM 3.5 inch SATA drive with a 3 Gb/s transfer rate.

Wait a minute. Why can’t you have both? What I mean is, why can’t a HDD be optimized for both performance AND capacity? In fact, some vendors do claim to offer just that i.e. HDDs that offer both best-in-class performance AND enterprise-level storage capacity. An example might be a 4 TB 7200 RPM 3.5 inch SAS drive with a 6 Gb/s transfer rate. But such “everything-optimized” drives generally don’t provide the same level of performance as performance-optimized drives. And while they may provide similar storage capacity to a capacity-optimized drive, they’ll usually cost more than capacity-optimized drives do.

So do you have to choose then between performance-optimized drives and capacity-optimized drives? Or can you just use “everything-optimized” drives for your servers? I’ll answer this question with a question. If your sweetheart has an expensive diamond necklace and some cheap costume jewelry, would you put them both in the same safe or jewelry box? Think about that for now, and I’ll return to this question soon in one of my upcoming articles here on WindowsNetworking.com.

Comparing enterprise-class drives

A picture is worth a thousand words, but a graph is worth ten thousand. There’s a simple way of visually comparing drives on the basis of performance, capacity, power and cost. To perform this comparison you need to understand something called IOPS (Input/output Operations Per Second) a common performance measurement used for benchmarking storage devices. IOPS can be measured in various scenarios such as reads vs. writes, sequential transfers vs. random transfers, the size of the block being transferred, and so on. For a basic overview of IOPS see this Wikipedia article, and for a much more detailed analysis see this article from Symantec.

The bottom line is that the IOPS for a particular storage device is best determined as follows:

Start by determining what the storage device will be used for. Will it be mostly sequential transfers or mainly random reads/writes? Will it be used for storing large files and documents, or for small transaction files?

Once you’ve decided upon the primary workload for the device, perform benchmarking tests on the storage device using an application like Iometer to measure the IOPS for the device for that particular workload. You need to do this because the IOPS will generally vary depending upon the type of data access scenario you’ve defined.

Now it would be nice if HDD vendors included the IOPS on the spec sheet for each model they offer (SSD vendors usually do this) but often they don’t. And it’s often hard to find published reviews of HDDs that provide measures of IOPS for different kinds of workloads. So for our calculations below we’ll simply use the following estimated values for the average IOPS for HDDs of different RPMs taken from several sources on the Internet:

RPM of HDD

Rough estimate of IOPS

5400

65

7200

85

10,000

135

15,000

190

 Table 1

Armed with the information in the above table (which of course shouldn’t replace real-world benchmark testing if you are planning to purchase large volumes of a particular HDD make/model) together with information usually found on HDD vendor websites and spec sheets, we can now calculate the following metrics for each make/model of HDD we want to compare:

  • Performance vs. Cost – How many IOPS will the drive deliver per dollar (or Euro etc.)?
  • Performance vs. Power – How many IOPS will the drive deliver per Watt of electrical power consumed?
  • Capacity vs. Cost – How many gigabytes of storage space will the drive provide per dollar?
  • Capacity vs. Power – How many gigabytes of storage space will the drive provide per Watt of electrical power?

Let’s now calculate the above metrics for two examples of real-world enterprise-class drives (vendor’s name withheld).

Example 1: Performance-optimized enterprise-class drive

Specifications:

  • 6 Gb/s SAS interface
  • 10,000 RPM
  • 32 MB buffer
  • 600 GB capacity
  • 2.5 inch form factor
  • 7.8 Watt power dissipation during read/write operations
  • Cost on NewEgg = $400

Metric

Ratio

Value

Performance vs. Cost

IOPS/$

135 IOPS / $400 = 0.34

Performance vs. Power

IOPS/Watt

135 IOPS / 7.8 Watt = 17

Capacity vs. Cost

GB/$

600 GB / $400 = 1.5

Capacity vs. Power

GB/Watt

600 GB / 7.8 Watt = 77

Table 2

Example 2: Capacity-optimized enterprise-class drive

Specifications:

  • 6 Gb/s SATA interface
  • 7,200 RPM
  • 64 MB buffer
  • 4 TB (4096 GB) capacity
  • 3.5 inch form factor
  • 9.6 Watt power dissipation during read/write operations
  • Cost on NewEgg = $480

Metric

Ratio

Value

Performance vs. Cost

IOPS/$

85 IOPS / $480 = 0.18

Performance vs. Power

IOPS/Watt

85 IOPS / 9.6 Watt = 8.9

Capacity vs. Cost

GB/$

4096 GB / $480 = 8.5

Capacity vs. Power

GB/Watt

4096 GB / 9.6 Watt = 430

 Table 3

Analysis

Now let’s represent the four metrics for each drive graphically so we can compare them more easily. To do this, we’ll use a graph that has the following axes:

  • Positive x-axis = GB/$
  • Negative x-axis = IOPS/$
  • Positive y-axis = IOPS/Watt
  • Negative y-axis = GB/Watt

Let’s now see what the metrics for the two drives look like when represented graphically in this fashion. Note that I’ve kept the scales on the two graphs identical to make comparing them easier.

Image
Figure 1: Graphical representation of metrics of a sample performance-optimized enterprise-class drive.

Image
Figure 2:
Graphical representation of metrics of a sample capacity-optimized enterprise-class drive.

These two graphs show us how different these two drives are when it comes to performance, capacity, power, and cost. For example, the second drive (the capacity-optimized one) not only delivers a much better deal in storage capacity, but also provides greater storage with less consumption of electrical power–something important if you’re deploying servers in a datacenter environment.

Conclusion

In my next few articles here on WindowsNetworking.com I’ll be examining this issue of choosing appropriate storage devices for servers in more detail. But let me leave you with this exercise. Look up the spec sheet for an enterprise-class SSD drive and calculate the four metrics defined above, then graph these metrics using a similar scale to the two graphs above. Now compare the SSD’s graph to the one above (Figure 1) for a performance-optimized enterprise-class HDD. How do they compare? When would using this SSD make sense in your server environment, and when would the HDD better serve your business needs for server storage?

About The Author

Leave a Comment

Your email address will not be published. Required fields are marked *

This site is protected by reCAPTCHA and the Google Privacy Policy and Terms of Service apply.

Scroll to Top