This content is part of the Essential Guide: Essential guide to hybrid flash arrays
News Stay informed about the latest enterprise technology news and product updates.

Installing SSD: Companies find pros and cons of doing it themselves

Installing SSD on your own without going through your storage vendor to save money? Here are lessons learned from people who have done it.

Taking a do-it-yourself approach with solid-state storage can be tempting. The price markups are often substantial with pretested or preconfigured drives from storage and server vendors. You can save money by purchasing solid-state drives on the open market and installing them yourself without going through your storage area network or server vendor.

Yet, although the do-it-yourself approach to a solid-state drive (SSD) install may be a feasible option, that doesn't mean it will make sense for most IT shops. More than a few IT consultants and industry analysts caution against the practice. An IT shop needs to do an assessment of its needs, application workloads and toleration for risk before installing SSDs on its own.

Installing SSD provides performance boost

Under the right circumstances, SSDs can supply the intended performance boost. For instance, Ecommerce Inc., a global hosting provider based in Columbus, Ohio, over the summer purchased two 480 GB Kingston SSDs online from Newegg Inc. for its corporate mail server. Since then, the IT staff has fielded fewer complaints about performance and suffered no outages due to hardware.

"This is a commodity and I [knew] what we needed," said Tim Perry, Ecommerce's chief technology officer.

Perry had already seen the benefits of SSDs in the Dell EqualLogic storage that Ecommerce's customers used for more than two years in scenarios where their IOPS needs were particularly demanding, especially when sending enormous quantities of email during the day. SSDs also helped Ecommerce with the boot process in blade servers.

He reasoned that SSDs could meet the IOPS demand for the corporate mail server, which services about 175 employees. Perry figured any potential downtime would have a negligible impact on service, and could be addressed quickly through restoring from backups and queuing mail traffic. Using SSDs in a RAID 1 configuration also gave him an added comfort level.

Perry was so pleased with the result that Ecommerce recently purchased three more Kingston SSDs for the database server that stores customer information. Two are for the database server in a RAID 1 configuration. The third is a spare for either the mail server or the customer database.

However, while Perry found the do-it-yourself SSD approach helpful at his company, he didn't necessarily recommend it for everyone.

"It depends entirely on the risk profile of the intended use," he said. "If the company can tolerate some risk, then buying retail may make sense. If not, then buying a vendor configured, tested, supported solution may be the best course."

Major university chooses not to install SSD

One major university has already shelved its do-it-yourself SSD experiment with its virtual desktop infrastructure (VDI). The IT staff had purchased 20 Vertex 4 Series SSDs from OCZ Technology Group Inc. to save money, and installed them in August in Dell blade servers, but removed the consumer-grade multi-level cell (MLC) SSDs in September after encountering unrelated technical issues.

Ed Swindelles, manager of advanced computing at the university (which he asked not be identified), said the school would probably not take a do-it-yourself approach again. He wrote in an email, "During troubleshooting [of] other problems, we had a lot of doubt about the problem symptoms due to the presence of the consumer drives, both internally and externally. They ended up not being the problem, but it was clear that we would have had an easier time with vendor-supported drives."

Prior to installing the Vertex SSDs, the university asked Dell if the third-party-supplied SSDs would affect support for the blade servers. Dell told the school it would continue to support the servers, but not the SSDs, according to Swindelles.

"When we ran into a problem; [the SSDs] were the first variable that Dell asked us to remove, since they were unsupported," Swindelles said.

Prior to the do-it-yourself experiment, the university employed Dell EqualLogic hybrid arrays with single-level cell (SLC) and enterprise multilevel cell (eMLC) SSDs and SAS disks for VDI storage. The storage system's automated tiering shifted most I/O operations to the SSDs during the VDI boot storms when students simultaneously logged in, Swindelles said.

The OCZ SSD trial was part of a plan to scale the school's VDI architecture from 700 to 1,000 concurrent sessions. The IT staff wanted to reserve the more expensive hybrid arrays for persistent sessions and use the less costly MLC SSDs in blade servers for non-persistent sessions. The setup called for eight servers with two Vertex SSDs apiece in a RAID 0 configuration for optimal performance.

Even though OCZ supplied a five-year warranty with the 512 GB MLC SSDs, the university set aside four of the 20 drives as spares to mitigate the risks associated with MLC, which has less endurance and reliability than SLC and eMLC. If one of the Vertex SSDs failed, the user session would shift to another machine in the VMware cluster so a staffer could take the node offline to replace the SSD, Swindelles noted.

"There is not a lot of data on quantifiable differences between MLC, eMLC and SLC. Rather, there's just a lot of marketing material about the advantages of the more expensive formats," Swindelles wrote. "So, we wanted to find out for ourselves if MLC would work in our use case."

The IT staff never got the chance to benchmark the MLC SSDs in the Dell PowerEdge M710HD blade servers. They merely noticed that they appeared to perform as well as the hybrid arrays with the SSDs and SAS disks. But, while fixing the non-SSD technical issues, they found it inconvenient to have all the non-persistent VDI sessions on local storage.

"Basically, every time we wanted to do performance maintenance on a VMware host, we had to either wait for all sessions on that host to close, or prematurely exit sessions," Swindelles said. "During this period of troubleshooting and rapid changes, we moved all of our VDI back to the SAN storage. The shared SAN storage allows for easy troubleshooting with VMware vMotion."

After resolving the technical issues, the IT staff never moved back to the local SSD storage in the blade servers. According to Swindelles, his team plans to continue to use and test the SSDs in development systems, but is unlikely to put them back into the production VDI environment.

"We need to evaluate the costs of the SAN storage versus the local SSDs. The hybrid SAN storage has been coming down in price and may offer enough advantage over the local SSDs to justify the premium," he said. "We are also less and less comfortable with the consumer SSDs and are going to evaluate the latest generation of eMLC SSDs when we get back to this."

The university worked with several vendors and consultants on the recent VDI project, and learned more about the distinctions between MLC, eMLC and SLC drives, including the ways in which they handle wear leveling and their expected endurance. IT staffers became concerned that performance might degrade sooner with MLC drives and eventually cause the drives to fail, according to Swindelles.

"In future efforts, we'll probably purchase eMLC," Swindelles said.

Based on his experiences, Swindelles advised enterprise users to do a thorough evaluation before purchasing SSDs directly from the manufacturer. "Unless you're a big organization with some margin for error, I'd recommend sticking with the solutions recommended by the vendor," he said.

Dig Deeper on Hybrid flash arrays

Start the conversation

Send me notifications when other members comment.

Please create a username to comment.