30% throughput on ethernet uk-ireland, 3 carriers
i have recently supplied a 100mb/s ethernet to my customer, between dulin - london.
3 telcos form this circuit end to end. i.e. 3 segements to the circuit.
when telcos run ethernet tests the throughput is ok.
however on live data ftp tests, the circuit consistently shows througput of c.25%.
I have a feeling that this is due to diiferent network translation problems.
Does anybody know what the problem could be here?
thanks
Software/Hardware used:
Software/Hardware used:
ASKED:
January 24, 2008 11:31 PM
UPDATED:
January 28, 2008 2:18 PM
RELATED QUESTIONS
- Layer 2 Ethernet throughput
- Load Balancing of Unequal Ethernet Bandwidth
- With both wired and wireless network connection, prefer to go wired, but metric doesn't always work.
- How do I show recruiters I'm valuable as a network analyst/engineer/manager when I have seven years of Telco experience?
- Calculating minimum throughput of a link
Answer Wiki:
My first guess would be that it's not a network issue but rather a hard drive issue. Try running several streams from several machines to several machines. For Example
SourceServer1 -> DestinationServer1
SoruceServer2 -> DestinationServer2
...
If you use 4 or 5 sets of servers you will probably get your full 100 Meg connection. In order to sustain read and write throughput above about 20-30 mb/s on a single server you'll need to have a very high end RAID array to handle the data.
To see all answers submitted to the Answer Wiki: View Answer History.
As a monitoring specialist, I see this as an interesting problem. I would suspect that the issue is not so much throughput as latency. Some simple tests would tell you, assuming ICMP is enabled:
I would start with traceroute through the link, looking at the latency of the various carriers. I’m a fan of the command line tools, but if you like GUIs, you might try a tool called Visualroute if you are on MS Windows (http://www.visualroute.com/), or Mtr on Linux/Unix (http://en.wikipedia.org/wiki/Mtr_%28My_traceroute%29). As the answer above states, it will be hard to separate disk I/O from network issues with an FTP test.
If you do see latency or packet loss with ICMP in a link, you can use that information to work with the carrier in question.
If latency is not the problem, you can watch the end point routers for late collisions and other packet errors. We use Cacti to do this, and set up graphs and thresholds to alert us when we do see bandwidth max out, or if packet errors start to show up. Errors that show up in these graphs speak to router overload and line quality issues.
If you have control of your end point routers, you might also use Cacti to look at the CPU and memory utilization on them, and make sure they are adequate for the job. Of course, it’s unlikely that you will get that kind of data from the intermediate routers, as they are managed by the carriers, but if you are able to measure what you control, you can at least narrow the problem down.
pathping is an excellent troubleshooting utility to look at latency & packet loss along a path.
Usage: pathping [-g host-list] [-h maximum_hops] [-i address] [-n]
[-p period] [-q num_queries] [-w timeout] [-P] [-R] [-T]
[-4] [-6] target_name
Options:
-g host-list Loose source route along host-list.
-h maximum_hops Maximum number of hops to search for target.
-i address Use the specified source address.
-n Do not resolve addresses to hostnames.
-p period Wait period milliseconds between pings.
-q num_queries Number of queries per hop.
-w timeout Wait timeout milliseconds for each reply.
-P Test for RSVP PATH connectivity.
-R Test if each hop is RSVP aware.
-T Test connectivity to each hop with Layer-2 priority tags.
-4 Force using IPv4.
-6 Force using IPv6.
It’s not a gui but it does give very valuable information within a single utility about the path between a source & destination.