langreiter.com plain, simple: comments-2006-03-20

LANGREITER.COM plain, simple

START

Montag, 20. März 2006

Guan Yang: "A lot of people actually underestimate how difficult it is to provision and back up large amounts of storage, even with tools like MogileFS; if we started coding 23 today we would seriously consider using [Amazon S3] for primary storage."

Guan also mentions S3's most serious weak spot: Compute cycles are far away. Being able to actually process the data on Amazon's machines would make the service a heckload more interesting (but sadly, at least an equal heckload more difficult to provide).

2006-03-20-s23

Olivier Travers has some gripes with AdWords/AdSense.

From the outside, it looks as if Google doesn't really care too much about AdSense (when it's so easy to monetize search traffic). What strikes me as especially weird is that they insist on doing their own content/ad matching (which certainly doesn't represent the pinnacle of targetting precision); only "premium" partners (like Findory) seem to get the opportunity to feed Google what they think are interesting topics for ads for a given visitor (see google_kw/[create google_hints]).

In any case, there are lots of chances left for alternate ad networks, which makes it all the more curious how long both Yahoo! and Microsoft take to ready their services for an international (or even non-beta) market.

Related reading (in German): smi's excellent run-down why there's the need for an eurocentric BlogAds-like network.

More related reading: Greg Linden, When AdSense goes bad.

2006-03-20-frustrad

Ruby Cheat Sheet

Ruby Hacking Guide (very partial translation)

2006-03-20-rooroo

gavin 6806 days AGO
If the compute cycles are far away, change what it is you compute to accomodate. Guan's example of video is a good one to refactor:

"[...] consider the example where users upload video files, but we need to do some post-processing on the video at a later date. This would involve downloading huge files from Amazon, doing the processing, and uploading them again [...]"

Downloading "huge files" isn't really the best use of something like S3 (or any sort of tuplespace-like thing for that matter.) Download individual frames of video (still big but not as big!) Sure you have to do a lot more but at least you're using the various parts in the stack appropriately and you could scale this sanely/predictably as well. Coordination of all this is another issue tho....

chris 6806 days AGO
I think the aggregate cost of communication is the issue here, and sadly that doesn't change (at least not for the better) when you're down/uploading tens of thousands of frames instead of one big phat chunk ...

gavin 6806 days AGO
If you bring "cost" into the picture, then nothing is fun.

chris 6806 days AGO
Heh.

guan 6795 days AGO
"service a heckload more interesting (but sadly, at least an equal heckload more difficult to provide)."

My understanding is that Amazon already does this with the Alexa Web Search Platform.

chris 6793 days AGO
which, predictably, isn't offered in the same fashion (sign up and go) as s3. I never got a reply to my inquiry (which, I have to admit, wasn't terribly specific).

Please log in (you may want to register first) to post comments!