[nSLUG] Recommendation for a graphical system monitor.
budman85 at eastlink.ca
Wed Jun 21 22:33:13 ADT 2006
I used MRTG years ago for monitoring network traffic on routers and
server loads. It uses perl and gnuplot for graphs.
Linux SNMP Network Management Tools
Server load is much different than network load.
Often times, CPUs are pretty much idle (except for NT). The bottlenecks
are usually between the keyboard and the monitor (j/k). They are usually
between network cards, backbone, and the WAN connection.
If your WAN connection(s) are 100Mb, then there can be a chance of
network backbone saturation (monitoring will help identify). If you have
more than one WAN connection that are heavily used, upgrading to a
Gigabit backbone would definitely help. A Gigabit backbone can maintain
roughly 10 times that of a 100Mb network. You can also look into
Etherchanneling and tunneling that can bump up the connection to 2Gb :)
Some improvements can be:
Enabling Full Duplex on the network interfaces if the router/switch
supports it. Try to do the same to the WAN interface, if it supports
it. You can see about 1/3rd performance increase. There are some
restrictions such as cable length and ethernet driver (as long as it
stable and not flaky). Most ethernet cards default to HD mode.
Check how busy the ports are. If it warrants it, upgrading the ethernet
cards being used for ftp to Gigabit may be worth it. Prices have dropped
over the years, so it can be an option.
Linux supports at least 16 ethernet cards (it may be more now in 2.4 or
2.6 kernels). NT can handle 4 NICs, but only 1 works. There is a bug..
umm.. programming infraction that only allows NetBios over NIC1. :)
This discovery led to me setting up Linux with 4 NICs and then
connecting to each of our subnets to do server backups to the Linux
Adding a Load Balancing device to the WAN, and then installing a few
Linux servers with multiple NIC's to handle the load.
Setting up the routing options in a custom kernel, enabling the QoS
features to make Linux simulate an ATM switch. You can assign priorities
to different services. This works nice, we saw major performance
increases by assigning higher priorities to FTP ports. You control who
gets the bandwidth.
On Tue, 2006-06-20 at 18:33 +0000, Stephanie Mayfield wrote:
> My boss is interested in seeing a graphical representation of the
> usage of the system resources on several *nix machines (RH enterprise
> 4 and CentOS 4.x). I'm thinking mostly CPU and IO, but the more
> information we can get (in one place) the better. Ideally it would
> span back a day or more so we can review.
> I'll summarize the business cases we want to use this monitoring tool
> to solve in case there are additional suggestions :)
> - When do we upgrade server x because it is no longer handling the
> load we wish to put on it.
> - Is server x suffering while server y is twiddling thumbs? (ie
> distribution of load across multiple machines)
> - Are there activity times on server x which are overloaded where
> re-scheduling is possible?
> The majority of the load consists of ftping files, mostly incoming
> although some are outgoing as well. Therefore I expect that I/O will
> be the main point of interest.
> Anyone have any recommendations? (Simple terms please - *nix newb
> here :)
> nSLUG mailing list
> nSLUG at nslug.ns.ca
Rich <budman85 at eastlink.ca>
More information about the nSLUG