3 ms·

The link is working again now, but has been replaced with a low bandwidth version. (The original version just contained additional images.) The amount of reques

The link is working again now, but has been replaced with a low bandwidth version. (The original version just contained additional images.) The amount of request generated by this site was unfortunately a bit too much for the web server. (I'm one of the server admins ... and have been in fear of a real DDOS attack during the last couple of minutes. ;) )

GoodIntentions 14y ago
If it isn't asking too much, would you mind telling HN about it? what kind of traffic you saw, what hardware/software platform you have in place?
- random123 14y ago
  The link got about 13000 hits so far, thereof about 5000 hits with a referrer value containing "ycombinator". But as only 3481 hits occured during the 3 days before it was posted on HN, I guess that actually about 9000 hits are due to HN. I think the hardware platform is not really interesting, as the DDOS was only caused by the worker pool of Apache2 being too small. Each hit of the site creates about 30 additional request for css,js,png and gif files. So 9000 hits in 3h mean something like 270.000 requests in 3h. Apparently this was too much for 50 workers.
  - GoodIntentions 14y ago
    so if my math is right approx 25 req/sec put paid to it. interesting. Perhaps I should re-think inlining js and css on landing pages.
  - fpp 14y ago
    That looks as if the majority of the hits went directly over to the google cache links mentioned here (otherwise most likely this would be > 50k hits at least). When the link was hitting the HN front page your site was already blocked - when the first person mentioned that the site was not reachable this HN post had about 60 points - anyway guess its the req/sec that killed you (and no cache - guess you don't have that much static content). Have you seen many facebook referrers from people who repost HN front page stories to Facebook?
    - random123 14y ago
      From the 23000 hits so far only about 300 came from faceboook ... not really that much. I'm pretty sure that the amount of req/sec caused the problems. After a configuration change between 150 and 200 apache2 workers were constantly occupied and neither network ,cpu or ram were even closely used to their capacity. (Before the change we had a hard limit of 50 workers.)

GoodIntentions 14y ago

If it isn't asking too much, would you mind telling HN about it? what kind of traffic you saw, what hardware/software platform you have in place?

random123 14y ago
The link got about 13000 hits so far, thereof about 5000 hits with a referrer value containing "ycombinator". But as only 3481 hits occured during the 3 days before it was posted on HN, I guess that actually about 9000 hits are due to HN. I think the hardware platform is not really interesting, as the DDOS was only caused by the worker pool of Apache2 being too small. Each hit of the site creates about 30 additional request for css,js,png and gif files. So 9000 hits in 3h mean something like 270.000 requests in 3h. Apparently this was too much for 50 workers.
- GoodIntentions 14y ago
  so if my math is right approx 25 req/sec put paid to it. interesting. Perhaps I should re-think inlining js and css on landing pages.
- fpp 14y ago
  That looks as if the majority of the hits went directly over to the google cache links mentioned here (otherwise most likely this would be > 50k hits at least). When the link was hitting the HN front page your site was already blocked - when the first person mentioned that the site was not reachable this HN post had about 60 points - anyway guess its the req/sec that killed you (and no cache - guess you don't have that much static content). Have you seen many facebook referrers from people who repost HN front page stories to Facebook?
  - random123 14y ago
    From the 23000 hits so far only about 300 came from faceboook ... not really that much. I'm pretty sure that the amount of req/sec caused the problems. After a configuration change between 150 and 200 apache2 workers were constantly occupied and neither network ,cpu or ram were even closely used to their capacity. (Before the change we had a hard limit of 50 workers.)