A year after introducing the Nexus 1000 virtual switch, Cisco is using that switch and its unified computing system servers as a proving point at VMworld in San Francisco. A Cisco-based implementation that taps into both EMC and NetApp SANs is powering the show's lab sessions. However, the setup performed so poorly in its first two days that many users were left unable to access the labs, they said.
Self-paced labs were all but completely offline until 2 p.m. on Monday, users reported, and a bootcamp session on Sunday, which demonstrated how to upgrade to VSphere, also had performance issues.
“Some of the more advanced stuff was doing host profiles, distributed switches, and you couldn’t do any of it. It just didn’t work. That was Sunday,” explained Jay Weinshenker, senior Oracle applications database administrator for Luminex Corp. in Austin, Texas.
As for the self-paced labs on Monday, Weinshenker noted that the performance problems made the experience “rather unpleasant. The labs are what I really look forward to. They’re good to get a lot of hands on experience. They talked to you a little bit and then you got to click around, but try actually messing around with some of it, and you couldn’t really,” he said.
Despite these issues, Weinshenker wasn't a seeking a refund on his conference. “I’m still getting a lot of knowledge out of it and the boot camp was free,” he said.
VMware acknowledged the issues, but emphasized that they were fixed by Tuesday. “We did have problems with one lab on Monday, but the issues have been addressed and all of the labs are fully functional. To give a sense of what we've been able to accomplish with the labs during the show, there have been 11,000 virtual machines that have been provisioned since 11:00 a.m. this morning,” said VMware spokesperson Mary Ann Gallo.
Cisco downplayed the performance problems and focused its message on the small footprint of the system, which it dubbed a virtual data center.
“This is one of the largest deployments of a single site virtual data center ever rolled out,” Ed Bugnion, vice president and CTO of Cisco's server access and virtualization business unit told Network World. “Cisco is bringing virtualization closer to the network, with direct visibility into the virtual machine. There are significant implications of virtualization on the network. The network was designed with static machines, static data centers in mind.”
There is nothing static about the Cisco Lab network. It isn't even hosted in a permanent room. Engineers from Cisco and VMware grabbed 1,700 square feet of hallway at the base of the escalators in Moscone Center and built the virtual data center there. They designed and built it in two months, with the last couple of servers, running bare metal hypervisors, provisioned in two hours.
The data center consists of 16 UCS systems total, running 512 blade, with a total of 1024 Intel Xeon 5540 Quad Core Processors (4096 cores) upon which it runs 776 ESX servers. It also runs Cisco UCS Manager. Each blade has 48 GB of memory upon which 40 virtual machines per blade are provisioned, five per core. The data center relies on two Nexus 7000 switches with two Cisco MDS 9500 switches connected to a multivendor SAN via 32 6120XP fabric interconnects. It uses both EMC and NetApp SANs, two units from each SAN vendor, hosting 37 terabytes of storage.
The whole shebang was expected to serve 4,000 online lab classes and 14,000 people over four days. Once the problems were fixed, at any given time, 1,500 users were accessing applications hosted by the data center causing a total of 20,000 virtual machines to run daily. And yet the data center consumes a mere 528 kilowatts, compared to the estimated 95,329 kilowatts that would have been required to build an equivalent physical version, Cisco says. It fits on 17,000 server racks, compared to the 154,000 server racks needed if it were using physical servers.
Users at the show said that they were impressed by the data center. One user said that he was heartened to see the multivendor SAN. Another said, “Makes me wonder if 10G is enough for what they're handling here.”
While frustration at the performance of the UCS system was initially high, others at the show were impressed with the efficiencies on display.
“I'm looking at the numbers. The potential cost savings and power savings are incredible,” said show attendee Navish Govindaraj, a director of product management for Informatica. He also added that the display is helping him conceptualize what an eventual multivendor cloud computing data center might look like.
VMware officials said that in the six years that the company has been doing VMWorld, the number of attendees for labs has doubled annually. It has been continually striving to increase capacity while maintaining a low, less power hungry footprint.