3 ms·

I guess we collectively blew it up =( This is why we can't have nice things...

Sami_Lehtinen 11y ago
I think they forgot rate limits. One of implementations were done using two actual routers and routing between those back and forth. Which means that it's really easy target to overload. I would have personally written light python script to fake it, instead of using hardware routers. It seems that pinging under 100 Kbit/s rate is enough to take it out of service, which is very little traffic. I would have liked to load test them up to double digit Gbit/s rate, but. This is disappointing. No challenge at all, single IP, no rate limits and poor capacity.
- davidcollantes 11y ago
  Would love to see it, how can I encourage you to?
  - Sami_Lehtinen 11y ago
    I'm busy with other projects, so I won't do it. But that's trivial. It would be a good practical learning opportunity if you're trying to learn about IP protocol and packet handling. Just use raw sockets on Linux, get data, and send responses. Alternatively you could use multiple IP addresses on same host, then you don't need to fake packets and can traditionally listen on ping on every IP. But that takes it to bit higher level and it isn't so fun anymore. Or you can set your computer to listen in on routed traffic, if you got router which will route subnet to your 'software' router. There are multiple ways to do it. Of course using two hardware routers is also ok, if you want to learn about configuring those. Different goal, different means. All of these means still require you to be able to control that subnet and set required reverse DNS information or run your own AS and rDNS.
- jlgaddis 11y ago
  If this is using a pair of low-end Cisco routers (as previous implementations did), it really doesn't take too many people doing this at once to "overload" them. Generating the responses is handled by the CPU (not the ASICs) and the processor in these devices is extremely underpowered, relative to, well, pretty much anything else with a processor in it. The processor in a Cisco 1841, for example, is an m68k at 250 MHz, if memory serves. Besides, I'm sure this was probably just something thrown together quickly for a good laugh and the responsible party likely didn't feel it was necessary to make this a highly available, fault-tolerant, geographically replicated service.
Sami_Lehtinen 11y ago
Also the latency between first hop of the story and last hop of the story tells about the system having very limited capacity. In case of the faking Script, delay difference between hops would be 0 ms. Now the numbers directly tell that there's plenty of latency and packet handling is taking a long time. -> Guaranteed to have a poor throughput & load characteristics.

I think they forgot rate limits. One of implementations were done using two actual routers and routing between those back and forth. Which means that it's really easy target to overload. I would have personally written light python script to fake it, instead of using hardware routers. It seems that pinging under 100 Kbit/s rate is enough to take it out of service, which is very little traffic. I would have liked to load test them up to double digit Gbit/s rate, but. This is disappointing. No challenge at all, single IP, no rate limits and poor capacity.

davidcollantes 11y ago
Would love to see it, how can I encourage you to?
- Sami_Lehtinen 11y ago
  I'm busy with other projects, so I won't do it. But that's trivial. It would be a good practical learning opportunity if you're trying to learn about IP protocol and packet handling. Just use raw sockets on Linux, get data, and send responses. Alternatively you could use multiple IP addresses on same host, then you don't need to fake packets and can traditionally listen on ping on every IP. But that takes it to bit higher level and it isn't so fun anymore. Or you can set your computer to listen in on routed traffic, if you got router which will route subnet to your 'software' router. There are multiple ways to do it. Of course using two hardware routers is also ok, if you want to learn about configuring those. Different goal, different means. All of these means still require you to be able to control that subnet and set required reverse DNS information or run your own AS and rDNS.
jlgaddis 11y ago
If this is using a pair of low-end Cisco routers (as previous implementations did), it really doesn't take too many people doing this at once to "overload" them. Generating the responses is handled by the CPU (not the ASICs) and the processor in these devices is extremely underpowered, relative to, well, pretty much anything else with a processor in it. The processor in a Cisco 1841, for example, is an m68k at 250 MHz, if memory serves. Besides, I'm sure this was probably just something thrown together quickly for a good laugh and the responsible party likely didn't feel it was necessary to make this a highly available, fault-tolerant, geographically replicated service.

Sami_Lehtinen 11y ago

Also the latency between first hop of the story and last hop of the story tells about the system having very limited capacity. In case of the faking Script, delay difference between hops would be 0 ms. Now the numbers directly tell that there's plenty of latency and packet handling is taking a long time. -> Guaranteed to have a poor throughput & load characteristics.