Smokepine is a very nice tool to have. We run ours on http://www.xssist.com/smokeping/
and it is publicly accessible. We want customers to know what they are paying for. A ping of 30ms during off peak
may be 200ms during peak to your particular location.
Web hosting companies have been offering uptime of 99.95%, regardless of size, regardless of infrastructure. Wow, anyone can offer great uptimes,
the Internet must be really reliable, right? Not really. The Internet is a best effort network. Google for smokeping targets, and look at the charts
available. Particularly for the 400 day chart, there's seldom solid, straight green lines.
How to interpret the graphs? Firstly, the host that the smokeping binary runs on does affect the results. We run smokeping on a pretty busy VPS.
There are times where there's lots of backup activity going on, lots of disk IO, and lots of bandwidth usage, and that shows up as spikes just
after midnight. Likewise, the choice of targets matters. Some of the targets are websites, and they do go down once in a while. Other targets
like routers give very low priority to ICMP, which is the right thing to do, especially when swamped with traffic, and the packets come back alot
slower than usual, although other traffic may not be affected. ping is a pretty good indicator of the health of the network otherwise.
One of the things that smokeping has caught is our main firewall slowing down packets. The firewall was configured to send out syslog messages, and
the CPU simply could not cope, and ended up slowing packets and dropping around 3%. It has also been very good in cases where customers from
specific countries report problems with the network. We can check back on SmokePing, and see if there's packet loss, and high latency during the
period reported.
01 Jan 2008
|