Second life: Revive your old workstation with SSDs

I’ve written on a number of occasions about the capabilities of enterprise-class SLC solid-state drives (SSDs) when leveraged in demanding primary storage environments. Their ability to both provide impressive sequential read/write performance and shoulder enormous OLTP loads has made them an almost indispensable component of high-end database and large VDI infrastructures.

However, to stop there ignores the enormous impact of consumer and workstation-grade MLC-based SSDs on the desktop market. The price of these devices has been falling consistently as their popularity has grown over the past few years (notwithstanding occasional spikes due to supply problems) and is poised to drop even further. Their low power requirements and ability to withstand vibration have made them a great choice for laptops, while their performance versus 5,400-rpm ATA disks has turned coveted sub-20-second POST-to-login boot times into a reality.

These lower-end SSDs have also found another interesting niche: in resurrecting older desktop/workstation hardware.

The project

Recently, I found myself contemplating a side project that would involve editing a sizable chunk of high-bit-rate 1080p/60 video. As an inveterate cheapskate, most of my workstation is a combination of recycled server hardware and older high-end rigs I was able to pick up at steep discounts, if I paid anything at all. As a consequence, none of the hardware I had was up for the task — at least not without some help.

The most likely candidate for the job was a five-year-old IBM Z-Pro 9228. When it was originally manufactured in 2006, this machine was almost literally the fists of God. With a pair of dual-core 3.0GHz Intel 5160 processors, 8GB of RAM, and a 15K SAS boot disk, it’s not terrible by today’s standards, though I shudder to think what it cost the original owner. Specs aside, would it be up to moving the bits necessary to deal with this kind of video? Probably not.

As I installed my editing suite and started importing video off the local disk, I could tell things weren’t going to be pretty. Even jogging through rendered video clips was a laggy, chunky affair, and it was extremely difficult to get any work done. I had already upgraded the machine with a capable CUDA-equipped Nvidia graphics board, so some of the effect work could be offloaded to those speedier multipurpose GPU cores. Still, the result was well below what I’d need.

The SAS bottleneck

I nearly decided to buy a newer machine, but curiosity got the best of me. In digging around in perfmon and doing some testing, I found that the four processor cores weren’t coming close to being fully utilized. Similarly, only a portion of the 6GB or 7GB available to applications was being used. That poor six-year-old SAS disk, on the other hand, was being murdered. (Anyone who’s owned a workstation with a 15K drive of this vintage knows it makes quite a racket — it’s hard to miss.)

Disk was definitely a big problem, but what could I do about it? One option would be to dig up a few more 15K SAS drives and throw together a high-performance RAID0 volume. But the idea of finding the right disks — much less trusting them — with a non-fault-tolerant RAID configuration wasn’t particularly appealing. Worse still, IBM, in its infinite wisdom, came up with a fancy toolless, quick-removal drive sled config for this machine, then didn’t think to include more than one sled. Adding more disks, regardless of what kind, would be kludgey at best. This is one of the many dangers of using proprietary workstation hardware.

The case for solid-state drives

It seemed clear that grabbing a decent-sized SSD was probably my best bet. However, the plan had its downsides. First off, though the system’s mainboard was equipped with a few SATA ports (the interface of choice for most consumer-grade SSDs), the system was old enough that they were first-gen SATA-I ports — there to support optical drives. Since most SSDs shipping these days are 6Gbps SATA-III or SAS, that meant I’d have to add a controller on top of getting a drive and figuring out how to mount it securely.

After some research, I decided to go a different route: getting a PCIe-based SSD. Having used these kinds of boards in the enterprise space, I’m familiar with how they work, but I was surprised by how cheaply I could get what turned out to be a very capable disk.

I ended up getting an open-box 120GB OCZ RevoDrive 3, which set me back about $320. At a little less than $3 per gigabyte, this is not the cheapest solution. However, if I was going to spend money, I wanted a piece of hardware that would remain relevant long enough that I could transplant it into whatever replaces the Z-Pro down the line. This card definitely fits the bill.

(Note: In my case, I used the RevoDrive as a simple data disk. If you decide to use this drive as a boot disk, you need to be very careful to ensure your system’s BIOS will be able to handle it. As I suspected, the Z-Pro’s BIOS didn’t have enough memory mapped to contain the RevoDrive’s option ROM, so that would not have worked in my case. Caveat emptor!)

Though it appears to the OS as a single 120GB (111GiB formatted) volume, it’s actually a pair of 60GB SSDs with their own controllers striped together by a third controller. Despite the comparatively limited overall performance potential of the Z-Pro, I was able to push sequential reads of over 400MBps, with sequential writes not far behind.

The published specs for this card are more than double what I was able to achieve, and I suspect that with a faster host machine, you could get much closer to them (I may test this at some point). Aside from having older processors, part of the problem was likely that the Z-Pro had PCIe 1.1 slots, while this is a PCIe 2.0 card — fully backward compatible, but you’ll get half the bus bandwidth.

The results are in

The effects were immediately noticeable. After porting my project over to the SSD and reconfiguring the project to use the SSD as scratch space, even flipping through unrendered content was noticeably smoother. Similarly, though rendering content took longer than I might want, doing so would easily peg all four processor cores — disk was no longer the bottleneck.

In the end, would I rather have a dual-quad Intel Sandy Bridge or AMD Bulldozer rig? Sure, but there’s no way I’d get that for the time or money I spent on buying and installing this one piece of hardware — not to mention the fact that I would have needed a similar SSD in whatever machine I might have bought.

The bottom line: Before you put old workstation equipment out to pasture, ask yourself whether a small helping of solid-state disk might cure what ails it. It may not be competitive with a brand-new machine, but it’s less expensive and could very well do the job.

This article, “Second life: Revive your old workstation with SSDs,” originally appeared at InfoWorld.com. Read more of Matt Prigge’s Information Overload blog and follow the latest developments in storage at InfoWorld.com. For the latest business technology news, follow InfoWorld.com on Twitter.