Archive for April, 2009

Dear Percona, please fork ASAP

Monday, April 20th, 2009

Oracle To Buy Sun For Approximately $7.4 Billion – Hold On To Your Hats .

If the past few years of history with InnoDB are any indication this is terrible news for MySQL.    Percona stepped up to solve that with XtraDB, now I’m hoping they take the lead on drizzle (which may mean forking and staff pilfering) and give us the drizzle+xtradb combo that we’d all love to see.

I wonder what they’d call it… PercizzleDB?

The difference between a feature request and a bug…

Sunday, April 19th, 2009

is whether you’re talking to a user or a developer.

A handy uptime and downtime conversion cheat sheet | Royal Pingdom

Friday, April 17th, 2009

A handy uptime and downtime conversion cheat sheet | Royal Pingdom.

 

 

 

Doesn’t show my personal favorite though, 99.95, or 20-minutes/month.  Its nice to work in the media industry.

Making sense of Dell’s M610 blade server memory configuration

Friday, April 17th, 2009

Over the last two weeks I’ve had a number of confrence calls with my Dell sales rep and some engineers about the new Xeon 5500 / “Nehalem” based blade servers, and specifically over the memory configurations thereof.  I have learned one thing for certain, this shit is impossible to explain without pictures.

The good news is, eventually I stumbeled across this great writeup with pictures on the Dell TechCenter Wiki: Nehalem and Memory Configurations

However its still a lot of words, and requires some intepretation to adapt to the M610, so with that in mind here is my explanation in ms-paint form:

Channels, Banks, and Slots

dellram-m610-basic

The Good Configs

Read the linked article for why, but the short of it is:

  • you want all your sticks of ram to be the same size
  • you want to use either 6, 8, or 12 slots at a time
  • you want to use 4GB or 8GB sticks.  Smaller would add up to less than 3GB-per-core, and given how much ass these kick at vmware that’d be a weird config.

Wait, what?  Show me

6 slots

dellram-m610-six

  • 6 x 4GB = 24GB
  • 6 x 8GB = 48GB
  • can run at 1333MHz or 1066MHz, uses all 3 channels per socket (best memory bandwidth option)

Eight slots

dellram-m610-eight

  • 8 x 4GB = 32GB
  • 8 x 8GB = 64GB
  • because two slots per channel are populated can only run up to 1066MHz

 

Twelve slots

dellram-m610-twelve

  • 12 x 4GB = 48GB
  • 12 x 8GB = 96GB

The Neckbeard Take? (sorry storagemojo)

6×8GB for 48GB is probably the sweet spot. It gives the best memory bandwidth performance and a 6GB-per-core ratio thats a good midrange for virtualization.

NOTE: HUGE CAVEAT! – 8GB sticks aren’t available for purchase yet. So if you plan on running vmware on these, you have a two month wait ahead of you.

Links

ran into some other good posts while I was at this, here’s the links

You can have multiple default routes in linux (centos/rhel5)

Thursday, April 16th, 2009

Just had to do a “live” default route cutover and while I was mostly sure it was gonna work its nice to have it definatively proven now.  You can add the second default gateway, then remove the first, so there’s never not a default route.  I’m not sure which one it kept using while they were both there, or if it per-packet loadbalances or what, but my ssh sessions didn’t even hiccup so at the very least it won’t knock you offline or interupt connections.   Of course this is assuming your upstream devices aren’t doing any nat/firewall stuff to complicate things.