I am having a particularly frustrating time with a client network. We upgraded some links from an old x86 routerboard platform running 5Ghz turbo and nstreme to an 802.11n system running NV2.

Previous system (turbo+nstreme):
Bandwidth Test: 65-70mbps
Windows File copy: 20mbps
application performance test (database client load): 4-5 seconds

New system (N + NV2)
Bandwidth Test: 85-100mbps
Windows file copy: 20mbps
Application performance: 10 seconds+

New system with MPLS:
Bandwidth Test: 95-130mbps
Windows file copy: 20mbps
Application performance: 45 seconds+

The application draws many hundreds of small files (couple KB each) as opposed to 1 big file. I cannot spot any issues with the setup. From what I can test the network should be performing outstandingly well, however actual performance is severely degraded from expected. Anyone have similar issues?

is this UDP or TCP test? copy files in windows is not the best way to test speed. Try btest (mikrotik) or iperf.

Still bad, try change frequency. Still bad, adjust adjust adjust antenna.

is this UDP or TCP test? copy files in windows is not the best way to test speed. Try btest (mikrotik) or iperf.

Still bad, try change frequency. Still bad, adjust adjust adjust antenna.

Sorry, I don't think you understand the problem. Wireless connection and bandwidth test is absolutely perfect. Up to 130mbps TCP transfer and 250+mbps UDP transfer. That is not the issue. The problem is that accessing a database application on the local LAN is lightning fast (1 sec to pull 200 records) but across the wireless network it is much slower (4-5 seconds with 5ghz-turbo, up to 15 seconds using NV2). This is despite the fact that the wireless network is running at wired ether speeds.

So I am saying the problem is somewhere else, perhaps there is an NV2 configuration I am missing that could optimize the link for transferring multiple tiny files, rather than for big file transfers.

I'll post a diagram of my layout for reference shortly.

is this UDP or TCP test? copy files in windows is not the best way to test speed. Try btest (mikrotik) or iperf.

Still bad, try change frequency. Still bad, adjust adjust adjust antenna.

Sorry, I don't think you understand the problem. Wireless connection and bandwidth test is absolutely perfect. Up to 130mbps TCP transfer and 250+mbps UDP transfer. That is not the issue. The problem is that accessing a database application on the local LAN is lightning fast (1 sec to pull 200 records) but across the wireless network it is much slower (4-5 seconds with 5ghz-turbo, up to 15 seconds using NV2). This is despite the fact that the wireless network is running at wired ether speeds.

So I am saying the problem is somewhere else, perhaps there is an NV2 configuration I am missing that could optimize the link for transferring multiple tiny files, rather than for big file transfers.

I'll post a diagram of my layout for reference shortly.

i still mean the same. test somewhere else. rather than radio-radio. try pc-radio-radio-rc and iperf. See results. I have manny times see a real different in testing radio-radio rather than some stuff behind.

Nv2 adds latency to link. its known. but try change to 20mhz bandwith (you may loose 10-20 Mb/s in speed) and focre to MCSxxx and only n.

Update: Seems the problem is with the RB1200 routers on ether side of the link! Replaced one of the routers with an RB1100AHx2 and response immediately improved hugely. I am running more tests now and will post back results.

Update II: I replaced the 3 x RB1200 with 1 RB100AHx2 and 2 x RB493AH (It's all I had in stock).

Performance increased massively and immediately. I can now do file copy across the network between 2 Windows 7 machines (using windows file transfer) at 96mbps, which is the limit on the 10/100 ether port of the 493.

The database application is another matter entirely, and the results are strangely inconsistent. On one of the links it performs better with NV2, and on a different link it performs way better with nstreme (NV2 is about 300% slower).

The only difference now is that one link uses RB435G's on either end (better with NV2 on) and the other runs 711G's (better with nstreme on). As a matter of interest the link running 711G's has better performance (100mbps vs 65mbps), but this is likely related to interference, not hardware.

I will be changing more hardware and shifting some things around to run more tests. First is to replace the 493AH with 1100AHx2 on 2 sites. This will remove the 100mb limitation. Then I will work on the slower wireless link to get that up to speed.

Just for interest: file copy across 2 wireless links runs at around 65mbps constant. I have attached the setup as an image for reference.

David,

Are the RB1200's by any chance using bonding?

We have seen very similar issues but only with bonding.

David,

Are the RB1200's by any chance using bonding?

We have seen very similar issues but only with bonding.

Nope, no bonding. It's a very standard routed network running OSPF. The issue is very weird indeed. When pinging from the 1200 router to a directly connected device (another router or a laptop) ping responses are sub 1ms. As soon as it goes through a switched network (tested on Dlink, HP Procurve and Dell managed Gigabit switches) ping responses to devices range from 1 - 20ms and sometimes higher. Just tracking this behaviour down took considerable time as we never suspected the router (we spent hours swapping switches and unplugging devices, thinking it was a faulty unit inside the local network).

As soon as we replaced the 1200 with a RB1100AHx2 or even a RB493AH the problems disappeared immediately. The irony is that we replaced perfectly working x86 based MikroTik routers with the 1200's because of compatibility problems with the new Mikrotik gigabit wireless boards.

Does Rb1200 gad latest version and latest bios version?

Does Rb1200 gad latest version and latest bios version?

yes and yes, first step I took was upgrading every device in the path

so now what do you do with such bad devices which are working hence not qualified for replacement or any other warranty issues??.....

becomes a capital liability.... does it?

so now what do you do with such bad devices which are working hence not qualified for replacement or any other warranty issues??.....

becomes a capital liability.... does it?

The problem only happens when going into a switched network. On highsites connecting to other routers it has no problems. Fortunately we can still use them elsewhere in our network.