5

We have network management system under linux, C/C++, perl and we need to test performance of this system. Is there a tool or way that would allow us to emulate 50 000 SNMP devices?

I don't know what more to say here... Please let me know if I should provide more information.

Any idea is appreciated.

Thank you Bogdan

Bogdan
  • 1,995
  • 2
  • 14
  • 12

5 Answers5

3

There are a few tools out there that will let you do that, however what I've seen is usually commercial software.

user39113
  • 297
  • 2
  • 4
3

You can try Raddle - it is open-source, written in Perl and based on honeyd, which should be able to emulate up to 65536 hosts.

L.R.
  • 977
  • 6
  • 22
2

Thank you for your answers. Here's the way we solve this problem right now:

  1. We have a linux VM up with 1 interface (eth0).
  2. We add 50'000 virtual interfaces (eth0:1, eth0:2 etc).
  3. An SNMP daemon is up and replies on requests through all the ip addresses.

This way we have 50K IP adresses in the network that reply to SNMP requests.

Though, I'm investigating Raddle. Maybe if it suits we'll switch to this solution.

Thanks, Bogdan

Bogdan
  • 1,995
  • 2
  • 14
  • 12
  • I have recently tried this approach, but it has a drawback: the SNMP agent (net-snmp) running on this machine provides information about all the virtual interfaces (in mib-2 ifTable, ipAddrTable, ipRouteTable,...), which can have some serious effects on the way the NMS handles these virtual hosts. So this approach is OK, as long as you do not use the above mentioned information to do some actions in NMS. – L.R. Mar 10 '10 at 19:34
  • You'll also run into performance problems since it's single-threaded. – Gambit Support Jan 06 '17 at 17:21
1

If the devices are sending traps, you could use Net-SNMP's snmptrap to simulate the effects of that many traps being sent.

DNS
  • 37,249
  • 18
  • 95
  • 132
0

The snmpsim tool reportedly can sustain the simulation of ~50K agents. It can simulate different agents by responding at different IPs or to distinct SNMPv1/v2c community names or SNMPv3 context names.

There are also the hints on performance optimization.

Ilya Etingof
  • 5,440
  • 1
  • 17
  • 21
  • All performance testing needs to consider many factors, including NMS-side power (eg. number of threads), query frequency, query makeyp (variables, type of request), simulator resources (CPUs, RAM, interfaces, ...), etc. Process context-switching is exceedingly expensive vs. threading. That's why there are commercial solutions for the most demanding simulations. – Gambit Support Aug 03 '17 at 14:45
  • Absolutely, snmpsim is not the fastest SNMP simulator ever. On the other hand, you can roll out as many instances of snmpsim as you need on the cloud for a performance testing phase of your product. I'm not arguing with @GambitSupport over performance merits, just noting the options we've got. ;-) – Ilya Etingof Aug 03 '17 at 15:34