Panasonic speeds its chip design process

When you’re transferring large files between the United States and Japan, you have to allow for a certain amount of delay, especially when your connection maxes out at 100Mbps. But the situation in which Panasonic Research & Development found itself was extreme, with many file transfers measured in hours, not minutes.

That all changed when the company implemented WAN acceleration appliances. Files that previously took hours now shoot overseas in a few minutes. Subsequent transfers of the same file may take just seconds, because only the changes get transferred, according to Iben Rodriguez, a consultant with Panasonic R&D.

Rodriguez, who works for the consulting firm Arcscale in Santa Clara, Calif., but spends most of his time with Panasonic, spoke about his WAN acceleration project at the recent Network World IT Roadmap event in Santa Clara and in a follow-up interview (available here via podcast). His implementation of SilverPeak WAN acceleration appliances sped up file transfers so dramatically that it required engineers in Japan and the United States to change the workflows they use to do business, taking what used to be a serial process and making it more interactive.

The Panasonic R&D unit where Rodriguez works, Panasonic Emerging Advanced RF Laboratory (PEARL), designs integrated circuits and has engineers in two main locations: the company headquarters in Santa Clara and in Japan. The process typically involves a design engineer coming up with an idea, which may be put into a large file with lots of calculations, then into a simulation environment. Engineers then design the circuits, lay them out on a chip and run multiple mathematical simulations to ensure the circuits perform correctly, with no interference.

Choking on gigabit files

For Panasonic R&D, each of those steps routinely involves files of 2GB or more going back and forth between the United States and Japan. The company had a 100Mbps VPN running between the two sites, but it still took hours to transfer the files, Rodriguez says. Engineers would often send files at the end of their day, leaving them for their overseas counterparts to pick up at the beginning of their next day, make changes, and send back.

In addition to being slow, that routine created problems with file versions, Rodriquez says. An engineer in Japan might be working on a day-old file while another in Santa Clara was working on a newer file. “Mistakes were made and we had some setbacks that we wanted to eliminate,” he says.

The company looked a number of potential solutions, including gigabit Internet connections in both Santa Clara and Japan, all of which were cost-prohibitive. That’s when Rodriguez started investigating WAN acceleration. After researching the various options, he decided the best fit would be an appliance that could cache files at each end and apply network acceleration and compression techniques. The device also had to have good security, including support for secure tunnels, encryption of files stored on its disk and a secure management interface. And it had to be able to handle terabytes of data going back and forth.

After looking at market share numbers and talking to references, Panasonic opted to do a proof of concept with just one vendor: Silver Peak Systems. That test went well, to put it mildly.

“We did some tests with a 400M FTP file. It reduced the transfer time from 49 minutes to two minutes,” Rodriquez says. “We didn’t understand how that was even possible. But we did some CRC checks on the remote files and they were fine.”

Another big benefit was that once a file is sent through the Silver Peak appliances, on subsequent transfers only the changes are transmitted. That was a big boon to Panasonic R&D, since often only small changes are made to its chip design files.

The company wound up purchasing three mid-tier Silver Peak NX-5500 appliances, each with 2TB of local disk cache, and installed them at its engineering sites in Santa Clara and in Japan as well as its server collocation site in Santa Clara. Implementation was “phenomenal,” Rodriguez says.

Now files go back and forth so quickly that it has forced engineers to change the way they work. It was actually disruptive at first because files that wouldn’t normally show up until the next day were there nearly immediately.

“The workflow used to be very compartmentalized. A verification team couldn’t start verifying a circuit that an engineer had developed until it had been through some other teams first, to make sure the layout was going to work or that the components were available,” he says. “Now we can be a lot more dynamic and more ad hoc. It really helps with the creativity because you can try out many different iterations and figure out what works best. So we can get a more finely tuned product.”

Words of advice

In terms of advice for others looking at WAN acceleration products, Rodriquez warns that the market is still young and changing rapidly. “It’s not clearly understood how the players will work out,” he says, noting it includes large players such as Cisco and Juniper as well as niche players such as Silver Peak. Given that, buyers have to do their homework and, ideally, conduct a pilot test. Failing that, ask for references, talk to them and even try to get real-world demonstrations.

Another bit of advice is to baseline your network before testing or implementing WAN acceleration. Panasonic R&D, for example, used Iperf to find out how fast its VPN tunnel was and at what point it would suffer data loss. The Linux-based Iperf, which is built into the Silver Peak appliance, enables you to throw an increasing number of packets at a link to find its saturation point. Panasonic R&D found a weak link in Japan that could handle only 3Mbps – this on a 100MB connection. The culprit was an older Cisco VPN router. “We got a loaner router with a TCP/IP offload card and then we were able to completely saturate the link at 100M,” he says.Finally, don’t always believe your application vendors when they tell you that WAN accelerators won’t work with their products, as one of Panasonic R&D’s vendors did. “We found that not to be true. Don’t be afraid to engage them and ask them to investigate it.”