Thread Rating:
  • 0 Vote(s) - 0 Average
  • 1
  • 2
  • 3
  • 4
  • 5
4.8.0.0.9047
#11
ninedd, I am crushed, but thanks for the report.

I have told you how I force the problem. Initially I wrote a program to run on the AP that stress tested it with a memory checker while a download was happening. I found I did not need to do that and instead have a HTTP server on my desktop with a 18 GB file that I download to my laptop that connects through a couple of war1b units. That system is on 2412 which it shares with 10+ other systems, and 3 of them are seen by the client at higher levels than its own AP. So, initially I had to do a sort of denial of service, but found it was not necessary.

The 4.9 series has never shown any issue. The 4.8 and previous all would stall and corrupt the transfer and the AP would need to be activated.

I am having a hard time with olsr on the 4.9 AP and on that test you setup for me the rate control goes crazy and drops to 1 mbps. Manual rate works just fine.

So, it looks like my initial elation at "fixing" the stall issue was short lived. I did fix and issue, and an important one, but obviously there is still a problem that your environment will trigger.

So ................. I am back to working on the 4.9 driver. I have to get the 1 mbps rate control issue fixed and get olsr working.

I am so sorry, but I am doing all that I can. I will not quit and this will get fixed.

Thanks for helping me.
Reply
#12
Again, it's not just a stall issue.
Reply
#13
OK, I believe we're getting hung up on semantics here.

The situation I am calling an AP Stall is where the clients start to slow down and possibly stop altogether. The only way to recover from this is an activate on the AP, and nothing needs to be changed, just an activate.

I believe this is what you are seeing. If it is not, then can you explain, again for me, what the situation is that you are watching for and manually activate to clear it.

If it was a car and started to slow down and then stop, I would say it had stalled. There are many different ways to stall a car, but the bottom line is that it is not going anywhere.
Reply
#14
I believe we may be talking about two separate issues here. At least it is for me. This is what I see.

1. Random reset on versions above 4.2.3. This may also be a lockup problem for some but it usually just resets itself frequently. There have also been some random slow down issues that a reboot seems to fix but this is not very frequent at 5GHz.

2. Performance degradation on many versions using 900MHz and 2.4GHz including StarV3. Depending on the software version this can be a quick performance slowdown from activate that ends up with less than half the bandwidth throughput anywhere from 15 seconds to several days. On V3 1.4 it happens immediately every single time after active, on 1.3 it can take up to a week. Doing an activate always brings it back. The only reason I mention these other old versions is because I'm wondering if the same thing is happening here. It appears that different software versions will see noise differently and make adjustments that it shouldn't be making due to it being temporary blips rather than a real noise floor. I do not see this issue at 5GHz probably because of less noise. When the issue happens I do not see the signal levels drop, just throughput goes to heck.

Hard to say if this is helpful or just clouding the issue but it really does seem like these are two unrelated problems.
-Andy
PTI Wireless
Reply
#15
Thanks for the feedback. It is appreciated.

Even though ninedd has reported that this does not fix the issue he is experiencing, I strongly believe that it does fix "something" that I can see here. This version has not had the stall issue in my environment and has allowed me to FTP a 18 GB file multiple times with no quitting or lockup or stall. In the older firmware this would trigger the stall issue and stop the download.

I am now working on getting the 4.9 series going. I have NEVER seen a stall and not even a FTP stop with the latest version. I have run the test on 2412 with 15 other AP's on that channel and a total of nearly 30 that can be seem.. Two issues remain:
1 OLSR does not like the wireless, but does refer all Ethernet routes.

In the 2. ninedd environment the rate control drops to 1 mbps almost immediately on a throughput test. The test runs great with a fixed rate so I need to stop the "dive to the bottom" nature of the rate control. I have never liked the rate controls because to me it seems just wrong to slow down way below the capacity because of noise or interference, especially interference. 1 mbps is useless and and with a few subscribers will very quickly have nothing but interference. The air time is so great that it allows more time for interfering signals. To my way of thinking, just do the retry as fast as possible and maybe let someone else have some air time. It is for this reason we all try and make sure that our clients have adequate signal to maintain a good rate.

So, to sum up, this release sure seems better than the ones before it, so I would that you give it a try on some spots that have been giving you issues. It has not created any bricks for me so I say it is quite safe and you can easily revert to the previous release if it is worse.



pti-andy Wrote:I believe we may be talking about two separate issues here. At least it is for me. This is what I see.

1. Random reset on versions above 4.2.3. This may also be a lockup problem for some but it usually just resets itself frequently. There have also been some random slow down issues that a reboot seems to fix but this is not very frequent at 5GHz.

2. Performance degradation on many versions using 900MHz and 2.4GHz including StarV3. Depending on the software version this can be a quick performance slowdown from activate that ends up with less than half the bandwidth throughput anywhere from 15 seconds to several days. On V3 1.4 it happens immediately every single time after active, on 1.3 it can take up to a week. Doing an activate always brings it back. The only reason I mention these other old versions is because I'm wondering if the same thing is happening here. It appears that different software versions will see noise differently and make adjustments that it shouldn't be making due to it being temporary blips rather than a real noise floor. I do not see this issue at 5GHz probably because of less noise. When the issue happens I do not see the signal levels drop, just throughput goes to heck.

Hard to say if this is helpful or just clouding the issue but it really does seem like these are two unrelated problems.
Reply
#16
I was not aware that it actually reduced the connection rate due to noise. I've never seen that happen but then again I'm using V3 on my 2.4GHz links and V4 on my 5GHz links so I really never see what ninedd is seeing. On my noisy links the total throughput continues to reduce (as if the rate had been reduced) from the moment of activation and stays low until the next activate. It never shows an actual reduction in the connected rate even though my throughput is almost nothing. This is with older software though so I can't say what is going on with V4 in regards to noisy 2.4GHz links.

I agree completely that the rate should only be reduced based on signal and Q%. When noise is present it is even more crucial that the most amount of data can get through in between the noise bursts. This means optimal data rate for the signal conditions not a reduction in rate.
-Andy
PTI Wireless
Reply
#17
The older versions used to have ability to set a limit on how low the tx rate would be allowed to go or was it a tx rate on how high it could go?? Either way maybe you could give us these options and we can figure out what works.
Reply
#18
Older versions had a cap on how high the rate was allowed to go.
StarOS Community Wiki: http://staros.tog.net/
Reply
#19
Further testing of the 5 GHz war1b and this release shows it is considerably faster than the 2.4 GHz version. As measured by Win10 TaskManager it is getting a solid 67 mbps on that 18 GB file download. The 2.4 GHz gets about 37 to 45 mbps from roughly the same locations. I just basically swapped the AP and client for the test.
Reply


Forum Jump:


Users browsing this thread: 1 Guest(s)