church of default

I was just telling someone last night that as awesome as EC2 is you still can’t run a high traffic LAMP site on it because the MySQL-write-master or “1000+ IOPS block device” answer is still missing.

Then this morning Vito passed me this: High Performance Computing Hits the Cloud
Which also led me here: New Amazon EC2 Instance Type – The Cluster Compute Instance

And for a few minutes there I thought they’d finally done it and I was wrong.

But alas, no, the “cluster compute instance” which appears to be one-half of a quad-socket/quad-7200rpm-drive server is offering you just that. 2x 7200rpm sata drives. Thats maybe 150 good random IOPS, not even close :(

You could string a bunch of them together in some kind of crazy LVM on iSCSI setup but you’d probably still only wind up with something in the ~600 IOPS family you can already get with almost-as-convoluted EBS setups.

Seems like there’s two ways around this currently. Keeping stuff in-house (well, colo) so you can scale up the hardware, or having your developers rewrite everything to not depend on a single MySQL-write-master so tightly. I always bet against the option that requires devs to redo stuff.

turns out yahoo has been building their own datastore thats sortof halfway between memcached and cassandra

this post is over a year old but I had never heard of it before
http://developer.yahoo.net/blog/archives/2009/06/sherpa.html

looks like they’ve been busy in the meantime:
http://developer.yahoo.net/blog/archives/2010/06/sherpa_update.html

no talk of plans to opensource it or even make it available as a service though

oops

Product Overview – Google Storage for Developers – Google Code.

Looks like the same basic premise, design, and pricing.     This is awesome.  Now you can solve the “what if s3 is down/slow” question with a redundant vendor instead of inhouse infrastructure.

Oil spill: BP had wrong diagram to close blowout preventer | McClatchy.

Choice quotes:

In the days after an oil well spun out of control in the Gulf of Mexico, BP engineers tried to activate a huge piece of underwater safety equipment but failed because the device had been so altered that diagrams BP got from the equipment’s owner didn’t match the supposedly failsafe device’s configuration
…

“When they investigated why their attempts failed to activate the bore ram,” Stupak said of BP engineers, “they learned that the device had been modified. A useless test ram _ not the variable bore ram _ had been connected to the socket that was supposed to activate the variable bore ram.”

“An entire day’s worth of precious time had been spent engaging rams that closed the wrong way.”

clearly not the cause, but there’s a DR lesson here for sure.