Boot drives are traditionally cheap, slow, low capacity drives that come with the server and have one job - to get the server up. While it may not seem like a critical function when a server’s running, and most servers are intended to run nonstop, the performance of the lowly boot drive can affect application uptime in a big way. Also, they’re often dedicated to boot and not used for other purposes, since most servers are connected to shared storage. For many servers, the boot drive is wasted storage space and a wasted drive slot.

Boot from SAN

One alternative to the boot drive is to boot from the SAN, if a SAN is available. But with more and more shared storage environments moving towards file instead of block storage, booting from a local drive is often the only option. Even when booting from the SAN is available, IT administrators often choose to boot from a local drive. It’s easier to set up, it’s often faster, it doesn’t add to network traffic and it’s a familiar process.

Performance of the boot drive is an often overlooked characteristic of production servers - until it’s needed. At these critical moments, the amount of time required to boot the system becomes critical indeed. If one considers when a reboot usually takes place the importance of a faster boot drive is more clear.

System down

When a server goes down unexpectedly it can be a fire drill. Sometimes an application hangs and forces a reboot, other times it’s a piece of networking hardware that requires that you ‘bounce’ the server. In either case, everything must wait for the boot drive to do its job. When there’s real troubleshooting needed to find the root cause of a downtime event, sometimes several reboots are required. Even with scheduled maintenance or software updates, time spent waiting for the boot drive is time lost.

A faster boot drive

If a local boot drive is a necessity the obvious question becomes “Is there a better alternative to the low-performance boot drives that are being used?” Solid state drives (SSD) may be the answer as they can offer some real advantages over the traditional boot disks that many environments are relying on. Using NAND flash technology, SSDs excel at read operations, and can provide orders of magnitude better performance, especially when fetching the small files that are typically read during the boot sequence. This can reduce the boot time in most servers from minutes to literally seconds.

A more reliable boot drive

With more data centers turning to high density blade servers and high density storage arrays, ambient temperatures are going up and with it, disk drive failure rates. Disk storage implementations rely on redundancy with RAID and secondary copies of data. But, since most aren’t redundant, the boot drive represents a single point of failure in a server. This means that the reliability of this relatively small storage device can have a disproportionate impact on the uptime of many applications. Since they don’t have any moving parts, solid state storage is generally more reliable than spinning disk, especially in higher density/temperature data centers, Using SSD as a boot drive can improve overall system uptime, compared with magnetic disk drives, especially when considering the potential for hard drive failure in today’s high density data centers.

What else can you get w/ SSD boot drives?

When a program requests data not currently stored in RAM it has to be fetched

from primary block-based storage and written to RAM before the application can use it. When the memory is full, the OS must decide which data or ‘pages’ currently in RAM can be moved to secondary storage in order to make room for the current data request.

This “page swapping” process essentially creates a virtual memory, as it allows the software applications to run effectively in a smaller physical memory space. But it can also add significant latency as operations running in memory are waiting for data to be swapped with (read from and written to) much slower hard drive storage.

Putting this swap space on SSDs can greatly speed up this process but manually configuring a server’s page swapping space can be complex. Page swapping with the boot drive is the default for most operating systems so using an SSD as a boot drive is a simple way to enjoy these benefits while enabling this resource to be better utilized as a part of day to day operations.

While not equal to DRAM used in computer memory, swapping pages to an SSD can be somewhat like a memory extension, at least from a latency perspective. In this way, SSDs can provide “RAM-like” performance at a much lower cost.

DIMM-based SSDs

An even better alternative may be to put NAND flash memory chips onto an industry standard DDR3 DIMM form-factor memory board, such as the Viking SATADIMM. When connected as a SATA III SSD this DIMM-based SSD can make the boot drive solution described above even more appealing. They plug into an open DIMM socket, like a standard DRAM memory module would, but they connect as a SATA device on the motherboard. These SATADIMMs supply from 30GB to 480GB of additional SSD capacity while not using any drive-bay real estate and a fraction of the power that a traditional 2.5” HDD does. Additionally, using DIMM-based SSDs allows you to use all drive bays for high capacity HDDs (i.e 2TB or 3TB), storing the less frequently accessed data.

Cache

Another use for this ‘free’ SSD capacity is as a high performance, ‘tier 0’ storage area or cache. PCIe cache cards are becoming a popular way to improve application performance quickly, by putting high speed storage capacity inside the server closest to the CPU. This storage can be used to accelerate critical applications, such as high transaction databases or as a read cache to hold commonly accessed data objects like indexes or metadata. But again, many low profile servers don’t have the space available to support a PCIe card. But they do have an open memory socket. A DIMM-based SATA SSD can provide high performance flash storage space to designate as a cache area.

Boot drives have performed the same historical function since client-server computing began. But using this legacy technology (slow disk drives) for what’s often a time-critical event (rebooting a server) doesn’t make sense, especially when it consumes valuable real estate in a low profile server. Replacing the anachronistic boot drive with an SSD to eliminate boot up latency would seem like a smart move. When this SSD can be plugged into an otherwise unused DIMM socket and provide additional benefits, like high speed memory swap space or a server-side cache, it might be a brilliant move.

“SATADIMM” is a trademark of Viking Modular Solutions

Related Articles

Viking is a client of Storage Switzerland       previous entry: How to Track the Cost of Storage