?

Log in

No account? Create an account
Operation Sled - brad's life — LiveJournal [entries|archive|friends|userinfo]
Brad Fitzpatrick

[ website | bradfitz.com ]
[ userinfo | livejournal userinfo ]
[ archive | journal archive ]

Operation Sled [Jan. 8th, 2004|08:44 pm]
Brad Fitzpatrick
[Tags|]

Desperate for food, I finally left my house in my sleigh (unchained front-wheel drive Honda Accord) in search of food.

From my previous bike ride (a comedic experience in itself) I knew that the main roads were pretty good (because I was biking in the car tracks), but my street was terrible.

To get off my street it was lots of thrusting and sliding, once having to reverse to get unstuck. Getting back in my garage felt like docking a boat.

But I'm back, and my belly is full.

I can't wait for the Great Flood of '04 tomorrow morning/afternoon. Should be hilarious.

I was reading in the newspaper about roofs collapsing, buildings burning down (because the firemen couldn't make it to the house), and trees falling over with the weight of ice, knocking out powerlines, etc. Chaos! Fun chaos. (Now that I have food in belly.)

Other accomplishments of the day: speeding up LJ! Been working on reducing CPU usage, since we're currently CPU-bound. Made some kick-ass improvements. And 4 new kick-ass web nodes are on their way too. That'll be cool.
LinkReply

Comments:
[User Picture]From: matthew
2004-01-09 05:29 am (UTC)

web nodes

Yes, buy more hardware ;)

How did you reduce CPU usage? We got the idea of using shared mem cache from you. Made a huge difference in performance under load.
(Reply) (Thread)
[User Picture]From: brad
2004-01-09 05:32 am (UTC)

Re: web nodes

CPU fixes were for the html cleaner:
http://www.livejournal.com/community/changelog/2004/01/08/

You using memcached, or something lame? :-)
(Reply) (Parent) (Thread)
[User Picture]From: matthew
2004-01-09 05:48 am (UTC)

Re: web nodes

well, according to that definition, it's something lame. :-)

we're using shared system memory to cache the results from SQL queries. Probably not the way of doing things, but it means that our 2.0GHz Celereon db server can handle more load than the web server can dish out :) It's not that we have a lot of data. The DB is only like 30M or something. It's the fact that we keep everything dynamic so with the cache that 30M turns into a few hundred megs of cache filling goodness.
(Reply) (Parent) (Thread)
[User Picture]From: brad
2004-01-09 06:23 am (UTC)

Re: web nodes

30MB is cached as a few hundred MB?

Sounds lame. :P
(Reply) (Parent) (Thread)
[User Picture]From: matthew
2004-01-09 03:34 pm (UTC)

Re: web nodes

yes. It's 30M in the DB, but when we start making SQL calls and caching the results we end up with lots of duplicate data looked at from different angles. Right now it's not a big deal since we have a gig of RAM on that system.
(Reply) (Parent) (Thread)
[User Picture]From: mart
2004-01-09 04:45 pm (UTC)

Re: web nodes

You could send the writes to disk but replicate to a slave which stores data on a RAM disk!

With only 30MB of data replication should be pretty quick, and if your data rarely changes it should be a good win!

(Reply) (Parent) (Thread)