Solid state devices and the hosting industry…

Shared web hosting and virtual hosting are 100% about resource management. From our perspective we want to maximize our hardware resources so that our customers have the cpu/memory and IO that they need while still having enough customers on a cluster of servers to make it cost effective to sell hosting at a reasonable price.

In my opinion CPU and memory constraints have mostly been solved in the hosting industry. Many hosting companies would disagree with me on that statement, but those that REALLY understand what the bottlenecks are that are hampering the hosting industry know that it is all about Disk IO and poor seek times.

Seek times on hard drives are BY FAR the #1 issue that Bluehost faces in terms of hardware/server performance. It astounds me that so many “techies” don’t get this when I talk to them about servers where hundreds of people shared the same storage either locally or through a NAS/SAN/SAS or other storage device. 90% of the high “CPU Loads” we see have nothing to do with CPU and everything to do with Disk IO blocking, and specifically the inability to seek fast enough to service all IO requests by the kernel. So what can you do?

The answer, in my opinion, is solid state drives. Hard drive technology has stagnated over the past 15 years as little has been done to increase actual disk speed. Sure, capacities have increased tremendously over that same period, but 3.7 ms seek time is still the fastest I have seen. I could buy drives with a 3.7ms seek time 10 years ago. Soild state drives – drives based on flash memory solve this problem. Most of these drives have a seek time of <1 ms, and many offer drives with seek time in the nano second range. These drives offer a tremendous speed improvement when the disk is under heavy IO load. We have implemented some of these devices in scenarios where MySQL databases were migrated to soild state devices, or where the “/” partition was completely overwhelmed by IO. The speed improvements were immediate and dramatic. MySQL was up to 50x faster in some cases and 5-10x speed improvement in database performance was common across the board. This is especially true in a shared hosting environment where many customers simply can’t optimize their databases for lack of time or knowledge.

Solid state drives are expensive and small. They only make sense when you are storing data that is accessed constantly, and that is reasonably small. 4 Gig to 64 Gig drives are common, but expensive. However, in the ultra competitive market of shared hosting every bit counts. You can count on Bluehost to start implementing these very soon as standard fare on all our servers. We have several in production now, but soon all our new servers will have these type of devices to yet again force the performance marker to go up one more notch.

Thanks,
Matt Heaton / President Bluehost.com

7 Responses to “Solid state devices and the hosting industry…”

  1. Flash drives have a finite number of write cycles which is why you don’t see them in servers. Is that an issue in your situation?

  2. Greg Schulz says:

    Matt you bring up some good and valid points in that for the most part server CPU processing and memory bandwidth continue to outpace disk I/O performance further increasing the server to storage capacity I/O gap (see the free industry trends and perspective white paper at http://www.storageio.com/Reports/StorageIO_WP_080706_Cover.pdf).

    One of the issues is that all too often a focus is on disk capacity instead of disk performance such as I/O operations per second (IOPS), bandwidth and latency. Consequently it’s not surprising to find organizations focusing on storage from a capacity standpoint instead of a performance standpoint resulting in some type or form of I/O bottlenecks.

    So where is the real I/O bottleneck in your environment? Are you using internal dedicated direct attached storage (DAS) relying on the disk drives in the servers, or, are you using a server with direct attached storage, RAID or non RAID as a general purpose file server? Or are you using RAID based purpose built storage arrays configured for performance vs. capacity being accessed via NAS NFS or CIFS file serving, HTTP or via block based access (SAS, SATA, SCSI, FC, iSCSI) via a web hosting server?

    Having been involved with solid state disk (SSD) devices as a customer, vendor and analyst I can attest to its strengths to address I/O and latency issues however there are certainly caveats and not the silver bullet for all I/O and disk performance related issues, that is, unless you have an unlimited budget which I’m sure your subscribers would not want to hear about. There are many different flavors of SSD including traditional RAM/SDRAM based as well as slower and lower cost FLASH based memory like what you find in USB thumb drives and hybrid disk drives like those from Samsung, Seagate, etc…

    RAM based SSD devices are available today in the 1TByte capacity range for .18 cents or less per I/O, likewise you can certainly find and deploy the smaller FLASH based devices being shipped in smaller servers and blade servers replacing traditional DAS (dedicated direct attached IDE/SATA/SCSI) storage devices.

    While SSD can address your I/O issues, keep in mind that there are also many very robust RAID based storage systems that when combined with SSD for your transaction, activity, event logs, high activity files or tables and other items, with the balance being placed on high speed RAID, either SAN (block) or NAS (NFS / CIFS) accessed you can find even more effective capabilities address and enabling better overall performance, availability, capacity and energy effectiveness. For a hosting environment like yours, there are even some more aggressive techniques coming to market depending on your specific current or desired architecture and topology to meet your service objectives.

    I would be happy to talk with you about SSD or other data and storage I/O infrastructure related topics when ever you like.

    Cheers
    GS

    Greg Schulz – Founder and Sr. Analyst
    StorageIO – IT infrastructure technology analysts and consultants
    http://www.storageio.com

    Learn about my book “Resilient Storage Networks” (Elsevier /ISBN 1555583113) at http://www.storageio.com

  3. Wes says:

    Anything that increases speed will be good. This is actually important when you have a site that performs a lot of tasks.

  4. Greg Schulz says:

    As an update a previous comment, solid state disk (SSD) technology continues to evolve both for pure DDR/RAM based SSD as well as so called NAND/FLASH based SSD and hybrid devices. For example Hybrid Hard Disk Drives (HHDD) that are either all flash, flash + ram, flash+ram+HDD are appearing in the market along with associated drivers. Likewise the SSD “storage systems” technology that is storage systems that combine FLASH and RAM or FLASH and RAM and HDD or RAM and HDD in a complete usable storage system some that include RAID and other data protection capabilities continue to evolve. For example Texas Memory Systems (TMS) recently announced a SSD storage system that scales up to 2TB capacity that is FLASH based with a RAM front end cache. The FLASH is RAID protected using an n+1 RAID5 stripe and parity scheme with the RAM being used to speed up performance. Seagate has a HHDD and Samsung among others have FLASH based hybrid drives.

    Needless to say as people seek out solutions to boost performance or reduce or maximize (do more with your existing power) your electrical power footprint SSD based technologies which tend to have about a 5-7 year popularity cycle are once again on the up swing of their cycle.

    Cheers
    GS

    Greg Schulz – Founder & Sr. Analyst – The StorageIO Group
    Author – “Resilient Storage Networks – Building Scalable Flexible Data Infrastructures” (Elsevier)

    http://www.storageio.com and http://www.greendatastorage.com

  5. Harley Witt says:

    Matt and others,

    NAND flash and solid state drives are fast at random reads and specific benchmarks will show that. However, even having 1 write for every 10 reads will bring the performance of an SSD to below a HDD. This is why people like Bongki Moon have written papers on how to optimize database performance by sacrificing more reads for much fewer writes. The day will come when true solution providers have systems with both SSD and HDD optimizing the best characteristics of both and then you will really see a great cost/performance benefit.
    Here is the paper I refer to http://www.cs.arizona.edu/~bkmoon/papers/sigmod07.pdf

  6. [...] Species List? SSD and Storage System Performance Are Hard Disk Drives (HDDs) getting too big? Solid state devices and the hosting industry Achieving Energy Efficiency using FLASH SSD Using SSD flash drives to boost [...]

  7. [...] Species List? SSD and Storage System Performance Are Hard Disk Drives (HDDs) getting too big? Solid state devices and the hosting industry Achieving Energy Efficiency using FLASH SSD Using SSD flash drives to boost [...]

Leave a Reply