Boston Scientific Corp.'s enchantment with all-flash arrays started with virtual desktop infrastructure (VDI),...
extended to I/O-intensive SQL Server databases and could spill into general-purpose use if prices fall enough.
The medical device maker, based in Marlborough, Mass., has four 10 TB and five 20 TB EMC XtremIO X-Brick units. Boston Scientific clusters 10 TB X-Bricks for its VDI environment and 20 TB X-Bricks with SQL database servers at its Minnesota-based cardiac rhythm management division, which makes implantable pacemakers and defibrillators. The most recent purchase -- a 20 TB X-Brick -- is pegged for test VDI and development servers at another location.
"All-flash arrays are really targeted to specific workloads right now," said Craig Englund, principal architect working out of the cardiac rhythm management division office in Arden Hills, Minnesota. "I'm looking forward to the day when the all-flash solutions come down in price enough where I can just put everything on them."
Englund said his only concern about using all-flash arrays was how long they would last given the fact that solid-state drives wear out over time. He said it helped that EMC increased the length of the flash endurance protection to seven years, and he became more comfortable with flash arrays as he gained a better understanding of how they work.
Inline deduplication and inline compression make all-flash arrays especially attractive for Englund over traditional arrays that do post-process deduplication. And he said he prefers all-flash over hybrid arrays with flash drives and spinning disks because he said he fears the hybrid performance might be unpredictable.
Boston Scientific needed predictable performance to address the challenge of high latency associated with patching and virus scanning with its VDI environment. The IT team uses full-clone desktops that are managed like a laptop provisioned to an employee, Englund noted.
Englund said he first heard about X-Bricks at VMworld in 2012. Boston Scientific was a longtime EMC customer, and the vendor provided a storage frame for a proof of concept the following year. Boston Scientific went into production with the X-Bricks shortly after the XtremIO product launch in November 2013.
Latency improved to about 0.3 to 0.4 milliseconds on the X-Brick cluster during normal production runs, according to Englund. On spinning disk, the latency had ranged from two to 10 milliseconds under typical conditions and soared beyond 20 milliseconds during the patching and antivirus scanning events, he recalled.
"Anything under 20 milliseconds is acceptable. People start noticing and having a poor user experience when it gets above that," Englund said. "We haven't had any issues with latency on the X-Brick."
XtremIO's X-Bricks enabled Boston Scientific to go from a rack of spinning disk running 700 desktops apiece down to a 10U clustered flash array using 10 TB X-Bricks, according to Englund.
"Those two storage arrays were way more in cost than the two bricks that I put in place," he said. "And it's performing tons better. Not only did I reduce my footprint, but I also get a much better user experience."
Compression will help, requires 'destructive' upgrade
Englund said he would like to put more virtual desktops on the same all-flash system eventually. He said beta testing of XtremIO's 3.0 upgrade, which EMC released today, indicates the team might see a 1.3-to-1 compression ratio with VDI.
That upgrade has brought EMC criticism for being disruptive because it forces customers to move data off the array and then move it back after the upgrade.
Boston Scientific has been testing the XtremIO 3.0 beta code on a loaner array from EMC. Englund called the upgrade "destructive … not just disruptive" but said upgrading his existing arrays is better than buying new hardware to get the compression.
"I'm in a highly virtualized world with VMware, so we can move things around storage-wise nondisruptively, even though it's inconvenient and takes some time," Englund said. "EMC has been really good at providing the swing capacity that we'll need to migrate off of our existing arrays to upgrade them to 3.0."
Testing the new inline compression capabilities with the company's SQL Server databases, which run as virtual machines (VMs), has produced a 1.9-to-1 capacity improvement, according to Englund. He said SQL Server data doesn't deduplicate well, but compression should enable the company to put more database servers on the flash array and save money.
One of the main draws for shifting the SQL Server databases from spinning disk to all-flash arrays was the need to load data faster to run reports. Englund said loading the data could take seven hours on a physical dedicated server hooked to enterprise-class storage. Moving the SQL workload to VM on shared infrastructure but connected to the X-Brick reduced the load time to two hours, he said.
"It's also given the application owners some confidence that a VM can perform as well or better than a physical machine," Englund said.
Englund said the all-flash arrays have enabled the IT team to take on larger and more critical and demanding workloads for the business. They have also paved the way for a new approach to disaster recovery in the case of the VDI environment.
Boston Scientific used to replicate its VDI environment between its Massachusetts and Minnesota sites when it used traditional disk-based arrays because new VMs took a long time and a lot of storage to spin up, Englund said.
"If there was a disaster and we needed to get hundreds of people working at another site, we couldn't have done that very quickly," he said. "But now with the flash arrays, I can spin up 200 desktops in an hour. So, we're leaving a small pool of desktops at each site for an initial disaster, and then we could spin up hundreds more to meet whatever demand there is."
EMC positions XtremIO for customers seeking scalability, low latency
Analyst cautions: Keep EMC XtremIO performance in perspective