Technology Toolbox

Your technology Sherpa for the Microsoft platform

Jeremy Jameson - Founder and Principal



Disk Benchmarks: SSD vs. "Raptor" vs. RAID

This post originally appeared on my MSDN blog:

Since I no longer work for Microsoft, I have copied it here in case that blog ever goes away.

Earlier this week, I posted about how I rebuilt my Windows 7 desktop while installing a new solid-state drive (SSD).

This morning, I thought I would share some performance numbers that I gathered from the various disk configurations that I currently have running in the "Jameson Datacenter" (a.k.a. my home lab).

Note that prior to installing the SSD in my desktop, I previously used a Western Digital 74 GB "VelociRaptor" drive as the primary disk in my desktop. In case you are not familiar with the "Raptor" drives, the key characteristic is the increased speed of the drive -- specifically 10,000 RPM. [A 74 GB hard drive is obviously rather puny these days, but keep in mind that I paid a premium for this drive ($189) back in 2005.]

A few years later, I swapped out some drives in one of my servers and added a couple of 250 GB drives to my desktop in a RAID 0 configuration. This yielded an additonal 500 GB of storage, which allowed me to replicate lots of content previously stored only on one of my servers and my Media Center in the family room (i.e. music, pictures, and video).

After adding the SSD this week, I now have three logical drives in my desktop (four physical drives).

To summarize, here's the current storage configuration for my desktop (WOLVERINE):

  • C: - 119 GB (formatted) - Crucial RealSSD 128GB (C300-CTF DDAC128MAG)
  • D: - 69 GB (formatted) - WD Raptor WD74 0GB-00FLC0 74GB
  • E: - 465 GB (formatted) - RAID 0 - ST3250410AS 250GB (x2)

Here's a pivot chart that I put together that shows the performance characteristics of each of these drives.

Disk performance on WOLVERINE
Figure 1: Disk performance on WOLVERINE

The chart above clearly illustrates why performance is such as relative term when talking about hardware. While the speed of the Raptor drive was quite impressive back when I bought it, it's almost laughable now when compared to what you can spend approximately $200 on these days (in terms of storage).

What I find most interesting about the above chart is that when it comes to write performance, a RAID 0 array comprised of some relatively old disks is actually faster than a brand new SSD. It's widely known that SSDs are much faster with read operations, than write operations -- but until I put together the above chart, I didn't truly understand the performance compared to other alternatives.

Even though the SSD is signficantly faster than my RAID 0 configuration with respect to reads (and almost as fast with respect to writes), it's not what I would recommend in most scenarios, primarily because of the high $/GB. [FYI, if I were to copy my music, pictures, and videos to my new SSD drive -- along with everything else I've already installed on the C: drive -- I would come up short by about 20 GB.]

To understand why I wouldn't recommend an SSD for most scenarios, take a look at the following chart which shows the performance of the SSD on my desktop (WOLVERINE) compared with other hard drives running on various servers.

Disk performance (various computers)
Figure 2: Disk performance (various computers)

In my mind, the RAID 10 configuration on BEAST is still the "sweet spot" in terms of performance, reliability, and cost. Keep in mind that you can probably put together a 1 TB RAID 10 configuration today at a lower cost than I spent on my 320 GB array many years ago. Heck, I bet you could probably put together a 600 GB RAID 10 array for about the same price as the SSD. [Of course, it's going to consume significantly more energy than an SSD, but that's a price I'm willing to pay.]

Also note that with the very large drives available today, you can get even better (or perhaps I should say "more consistent") performance by using "short stroking." However, if I were to go down that route for my home lab, I'd really have to get my head examined.


  1. # re: Disk Benchmarks: SSD vs. "Raptor" vs. RAID

    March 11, 2011 7:56 PM

    I have the Crucial RealSSD 256GB CTFDDAC256MAG-1G1 in my Lenovo W510 laptop. It's said to be 30% faster than the 128GB version. In fact it's so fast, it's my SATA II controller in the laptop that's the bottleneck. Some tests promise another 30% gain if I had a SATA III controller. Maybe writes are still a bit slow compared to reads. But I'm telling you - being able to boot directly into my SP2010 VHD developent environment from the boot manager and almost be working "on the metal" is like sitting on a blazing nuclear missile. I happily give up gorillians of megabytes in exchange for the speed I get. Just thought I'd say. :-)

    Btw, great blog - I'm an avid reader. Keep up the good work!

  2. # re: Disk Benchmarks: SSD vs. "Raptor" vs. RAID

    March 13, 2011 5:14 PM
    Jeremy Jameson

    Peter, I hear ya.

    I wish my desktop had a motherboard that supported SATA III. According to the stats on newegg, read operations on my SSD would be almost 34% faster if I moved from SATA II to SATA III. Wow!

    To be honest, I haven't yet attempted the "boot from VHD" approach with my laptop yet. The biggest drawback I see is not having access to things like Outlook, My Documents, etc. In Microsoft Consulting Services, we use to get two laptops (one reasonably new laptop and usually somebody else's hand-me-down) so we could run separate environments for scenarios like this. However, the days of lugging around two laptops -- at least for me -- are long gone.

    I also am not a fan of running "everything on one box" -- like the SharePoint "Contoso" images provided by Microsoft. I'd much rather separate the domain controller onto a separate (virtual) machine. I think it's because over the years I've simply encountered too many obscure issues that occur when one of our products gets installed on a DC.

    The other thing I'm really starting to use more and more often -- especially with SharePoint 2010  -- are Hyper-V snapshots. I used to avoid differencing disks in VMs like the plague (due to performance issues), but I was pleasantly surprised a few months ago to find that Hyper-V snapshots actually perform quite well. In fact, my primary SharePoint 2010 VM currently has two snapshots so I can quickly rollback to a couple of key points in time (which is probably something I should cover in a separate post).

    If I do ever end up having to run SharePoint 2010 on my laptop (and I need more features than are available when running directly in Windows 7) then, to be honest, I'll just continue to run VirtualBox. Yeah, I know, it blows that this is now owned by Oracle, but oh well. [This "blows" because I'm a Microsoft employee, not because it's inherently bad that VirtualBox was bought by Sun -- which is now owned by Oracle.] I doubt my manager would approve an expense report for VMWare Workstation -- not that he should have to anyway.

    I keep thinking one of these days Microsoft will announce a version of Virtual PC that will support 64-bit VMs. Okay, enough ranting...I need to move on to more productive things ;-)

Add Comment

Optional, but recommended (especially if you have a Gravatar). Note that your email address will not appear with your comment.
If URL is specified, it will be included as a link with your name.

To prevent spam from being submitted, please select the following fruit: Grapes

Please add 7 and 6 and type the answer here: