Here's the e-mail i got from dreamhost support:
I'm really sorry but I had to disable the website XYZ. I'm
not sure what you had running with that server, but it was jamming up
networking for the entire server, which was causing NFS to fail, and thus
was driving up load on the entire server. As soon as I disabled the
website the load came down and the server was able to talk to the file
I apologize for the inconvenience, but when a website breaks a server
this badly, there's no way we can re-enable it.
Thank you for your understanding, and again I'm so very sorry about this.
That website servers up large files also (40 to 350mb) which might be cause of
the problem, seeing all drives are networked at you guys (a weakness i do here
I'll look into it, but i'm disappointed on the weak solution you guys apparently
have. Did you check was it all I/O wait? Seen the same happen here too, tho
those times, just too slow HDD. (Checkout SAR for i/o monitoring, CentOS does
have package... package name i cannot remember)
- Aleksi "
with SAR i was meaning iostat, remembered the name wrong
and yes that one mention is intentionally Joda speak.
When we notice a site crashing a machine, we have no choice but to
disable it. We cannot allow one user to overconsume system resources and
bring down a machine. It's not fair to other user on the machine. If you
have any additional questions please let us know.
The 2nd reply is completely legitimate, honest, and outlines good support, he mentioned the problem, action and why.
Now, the thing here is that they are denying usage of what they give to me, because of bottlenecks in their infrastructure (too slow connection to network storage, and/or network storage device too slow to handle that many concurrent users and/or websites).
It's OK imho that they disabled the website, infact, i would prefer that way if there's a real problem. What they are doing although, is denying the right to use it to the full degree. Seeing as the most traffic consuming part is static content.
Here are google analytics statistics for that site:
8th day: 5,962 pageviews and 2,065 visitors
9th day: 2,981 pageviews and 846 visitors
10th day: 2,091 pageviews and 502 visitors
11th day: 52pageviews, 36visitors ** WEBSITE DISABLED **
DH Panel BW usage, 01/10 to 11/10: 145321.2 MB
So i remembered incorrectly, it was about ~140Gb for that period.
Not that much really. Cycle estimate: 396330.545 MB or ~394Gb
and as you notice, they clearly state that network was bogged down.
So basicly, that little amount of traffic got their server jammed :O
Quite little, eh? another intresting thing is that the spike was dimming already and everything getting back to normal.
I would think that's rather intresting, seeing that clearly it will use less than 1/4th of the assigned bandwidth allotment.
Now, i know for a fact, that any of my servers could handle that kind of spike, with increased loads, yes, but decreased responsiveness? Highly doubtfull.
Btw, linux load means just running processes, not real usage, but it's a very good indicator.