Slashdot Mirror
A Look At One of Blizzard's Retired World of Warcraft Servers
MojoKid writes "At last count, Activision Blizzard pegged the number of World of Warcraft subscribers at 10.2 million. It takes a massive amount of gear to host all the different game worlds, or realms, as they're referred to. Each realm is hosted on its own server, and in late 2011, Activision Blizzard began auctioning off retired server blades from the days of yore to benefit the St. Jude Children's Research Hospital. They sold around 2,000 retired Hewlett-Packard p-Class server blades on eBay and donated 100 percent of the proceeds (minus auction expenses) to the St. Jude Children's Research Hospital, which seeks to advance the treatment and prevention of catastrophic diseases in children. This article has a look at one of those retired server blades."
For the Horde, I mean, FOR THE CHILDREN!
With shipping, which was almost as much as the server itself, I paid $243.50 for this showpiece.
Hmmm $100 or so to ship? Someone's padding that expense line. I would not flinch at $25 to $50 but this smells of those ebay auctions where its $0.01 for the product and $50 to ship.
"Science flies us to the moon. Religion flies us into buildings." - Victor Stenger
It looks like any other blade, once you ignore the marketing decals put on it.
For large sets, this will be our guide even unto death, for the LORD will work for each type of data it is applied to...
vmware still has issues with all the I/O going through the hypervisor. the blades have local storage for the OS
there are still applications like Cognos and others that say that if you run them in vmware then use a separate physical server due to the I/O demands and the fact that you have to specially code around the oversubscribing feature of vmware
Um, where do you think those Xeons and Opterons are installed? In individual towers? 1U servers are basically the same as blades except you have a lot more smaller redundant parts (power supplies, fans, etc...).
Plus, you are griping about hardware that has been retired.
I read the internet for the articles.
Can't speak for all RAID cards, but the ones I've worked have a certain amount of RAM soldered onto the card and a slot for additional RAM that's semi-optional (the last card I worked with required additional RAM to add another drive to the RAID 5, but was working fine with just the onboard RAM before that). So it may not be missing it so much as never having had it in the first place.
Hipster IT admins, ASSEMBLE!
"Blades are so mainstream. People in the know use a CPU with buttloads of cores. I'd tell you what brand, but it wouldn't matter, you've never heard of it."
"Sure, 10-speeds save physical space, but they add complexity, cost, points of failure, and the heat they generate is the same. That's why I prefer fixies."
"Why mess with blades? You can't even put a bird on them."
The article writer doesn't mention the specs of the blade, isn't interested in knowing if it works and thinks its ugly?! He has no interest in server tech or playing wow. Why waste our time linking to this article?
The main saving grace of the humble 1U is that it doesn't have a vendor who has you by the balls for the next 14-ish systems you buy, along with a variety of option cards and things. Your basic rack doesn't provide much in the way of amenities, leading to lots of messy duplication of 40mm jet-fans and PSUs and a cable mess; but it just doesn't have the lock-in of a physically and logically proprietary cardcage...
So far, the blade guys have had a difficult time not pocketing as much of the extra efficiency value as they can, while the commodity 1U knife-fight is wasteful; but it is rather harder for your vendor to achieve market power over you.
Um, where do you think those Xeons and Opterons are installed? In individual towers? 1U servers are basically the same as blades except you have a lot more smaller redundant parts (power supplies, fans, etc...).
Plus, you are griping about hardware that has been retired.
1U servers are basically not the same as blades, lol.
It only makes sense to employ people if you have a job for them to do. If Blizzard had nothing useful for them to do, keeping them around to twiddle their thumbs doesn't make much sense.
I get the feeling their backend design wasn't the best. For years they took their servers down every single week for a massive 6-8 hour maintenance period. This wasn't for updates, this was just regular. Patches took forEVER to happen. It clearly wasn't something like "Take things down, roll out new code, run checks, bring it online." Given that some things would only affect particular realms it was pretty clear they were doing things like running series of scripts and commands to upgrade things, and the process shad trouble in certain configurations and so on.
So it wouldn't surprise me if they did things like store data on the blades themselves and so on. I can't say for sure, since Blizzard has been secretive to the point of paranoia about how things work on the back end, but my experience with the game leads me to believe they did not have a particularly good backend setup.
Blades are all about density. If I can squeeze 10 dual-host blades in a 7U enclosure, that's 13U saved vs 20 1U servers. Add the fact that many modern blade enclosures integrate modular switches, and you can squeeze 120 hosts per rack, instead of just 38-40. The hardware cost difference is negligible, since you're buying one set of redundant power supplies to power all 10 blades. The enclosure itself is costly, but the blades aren't much pricier than a comparable server board.
If you're deploying lots of them like Blizzard, choosing blades means you only need 1/3rd of the floor space, 1/3rd the shipping cost, 1/3rd the installation labour, which represents a huge chunk of change when you're colocating at top-tier datacenters all around the world.
Blades may not make sense for everyone, but don't write them off just because your needs are satisfied by simpler solutions. Virtualization is a great tool that offers tremendous flexibility and reduced costs, but it is not a magic bullet to solve every problem. It excels at handling small jobs, and fails hard with large ones. For example, virtualization struggles with I/O heavy workloads, which are becoming increasingly important with the meteoric rise of data warehousing and distributed computing. Processors are the easiest part of the equation.
-Billco, Fnarg.com
Because you can treat a single blade chassis as a single computer, despite the fact that it has 10+ computers in it. So, rather than separate boxes tied into a SAN, you have a single "computer" with directly attached drives (SCSI drive farm) for better performance. Then you cluster piles of those with a shared SAN for what must be shared across them. Better performance than separate machines.
Oh, and blade servers have better reliability, even if you think they have more points of failure. And, depending on your setup, space is a cost that is a consideration, and compactness will save money.
Learn to love Alaska
No they are not still using blade servers, hence the reason they are selling them off as historical pieces of art basically..
"But this one goes to 11!"
I looked at the auctions when they originally occured
they were shipped INDIVIDUALLY from France, so the price ain't that bad
Now, WHY TF they weren't shipped to one point in the US (say Blizzard headquarters) and then individually shipped to buyers-- escapes me.
every day http://en.wikipedia.org/wiki/Special:Random
It only makes sense to employ people if you have a job for them to do. If Blizzard had nothing useful for them to do, keeping them around to twiddle their thumbs doesn't make much sense.
This.
I play paper and pencil games with someone who had his department basically cut in half.
Over the past few months a lot of their tasks were made more automated and they were being sent home early due to a lack of work.
Apparently there was a cost/productivity metric that was calculated for each of them and the more expensive ones were let go.
And according to him, the severance packages were nice enough that it was clear that this was not a 'we can't afford you any more' type situation.
Character data would be stored in a database (in Blizzard's case, Oracle). The local drives on the blades would have game data and server executables, which would be even more valuable than character data to the right people (gray-sharders, botters, and other nefarious types).
VMware could handle hundreds of thousands of IOPS into a single host years ago. Pretty sure it's over a million now.
Outside of exceptionally unusual corner cases, if your storage system can handle it, VMware is not going to be a bottleneck.