Not even a miniature tsunami could stop the InfoWorld blade server shoot-out of 2010. But it sure seemed like a possibility when we were running for high ground at 6 a.m. on a Saturday morning with nothing but the clothes on our backs and the sounds of warning klaxons reverberating throughout the entire island. Luckily, we wound up eating pancakes and getting bored watching CNN from lounge chairs rather than avoiding killer tidal waves.
In fact, it looked more likely that logistics would play the biggest role in whether or not this test happened at all. Multiday shipping delays and several near misses while wrangling racks of gear all contributed to the concern that maybe this test just wasn’t meant to happen. There's a certain feeling you get in your gut when you witness a half-ton rack of gear nearly topple over on a freight truck liftgate. It's closely related to the sensation you experience when you watch the same rack rocketing toward the back of the freight truck dragging the driver behind it.
Despite more near misses than I want to count and after a mountain of effort by many people, the 2010 InfoWorld blade shoot-out posts on InfoWorld.com today. It's the result of months of preparation and grueling weeks of heads-down work. How it all happened is a story unto itself.
It was something of a miracle that it transpired at all. Due to the economy, hardware availability, lab availability, and a host of other reasons, the date of this test kept getting pushed back. Part of the problem was the delayed release of Intel's Westmere CPU -- which turned out to be a good thing in the end. HP, IBM, and Dell all showed up with pre-release Westmere blades, so the hardware we tested was absolutely cutting-edge, representing the best blade hardware available today.
Sun was originally invited to the test, but then Oracle gutted Sun's PR team, so nobody who answered the phone knew anything about anything. It was sad; though getting Sun would have been the icing on the cake, the cake was still quite tasty with titans HP, IBM, and Dell onboard.
All the testing was conducted at the ANCL (Advanced Network Computing Laboratory), at the University of Hawaii. The weekend before the tests started, major earthquakes hit Chile and spawned tsunami warnings on Hawaii. The plan had been to spend that Saturday in the lab complete the preparations for the tests, including doing dry runs with Ixia's IxChariot and IxLoad testing tools. That was not to be. Instead of a day of preparation, we ate pancakes in Brian Chee's living room and watched CNN breathlessly predicting the demise of the state of Hawaii. Hawaii survived, but Saturday was lost; we spent Sunday in the lab instead.
Once all the parts were in place, we had a very tight schedule and a series of tests to run on each big box as it landed in the lab. There wasn't much room for error -- possibly a half-day per vendor that could be stretched a bit if something really bad happened. For example, a 1,000-pound crate full of blades could somehow take an extra five days to reach the lab.
I'm not 100 percent sure of the reasons why, but HP's chassis almost didn't make it in time. As it turns out it's a good thing it did arrive, because HP's blade wound up the winner, though it was touch-and-go for a while. What was supposed to be a two-day shipping time stretched out to seven days and caused ulcers to flare across six time zones.
It finally showed up at the last possible minute, and just to increase the drama, it nearly fell off the back of the freight truck liftgate while being delivered. Five guys including myself were standing around the back of that truck, watching the delivery drivers roll it out. Apparently, the liftgate was rated for only 750 pounds or so, and the half-ton rack caused it to descend early, though part of it was still in the truck. The whole thing started tipping over. All of us jumped forward, pushed on the crate, and averted disaster, although the rack landed rather hard. In the end, there was no damage and the only problem was one blade that popped a single DIMM from its socket. It was nothing short of a miracle all around.
The arrival of Dell's hardware was more like an old-fashioned "Star Trek" beam-down. Dell overnighted the chassis, expecting it to take at least two days for the 600-pound rack to get to the lab. In fact, it traveled 5,000 miles and arrived the same day. Unlike HP, Dell's team shipped with FedEx and apparently made the cutoff, which resulted in the rack arriving 18 hours later -- a truly amazing display of efficiency.
Testing the Dell solution was also highly efficient. Everything went as smoothly as possible, and we were able to knock off a bit early both days, with all the relevant benchmarks run, all the data gathered. It was somewhat strange that Dell sent only a single tech with its gear. Whether that decision was out of hubris or cost savings, the tests ran with flying colors anyway.
IBM, on the other hand, had issues. It had trouble sourcing some parts for its chassis and needed more time to prepare. As a result, IBM missed its original dates, and the rescheduling ate up the two contingency days at the end of the test schedule before the tests even began. Luckily, the gear arrived the evening before the rescheduled test dates and testing went very smoothly. By Friday evening, the last day of the test, we finally felt that we were going to get everything done in time. It was all the way down to the wire.
The whole experience was a weird combination of a roller-coaster ride and forced march, full of suspense and predictions of imminent disaster. But we made it. The result is the 2010 InfoWorld blades shoot-out, a good hard look at the best blade systems available today. Enjoy.
This story, "Tsunamis and falling crates: Behind the blade server review," was originally published at InfoWorld.com. Read more of Paul Venezia's The Deep End blog and follow the latest developments in servers, processors, and other hardware at InfoWorld.com.