Opened 4 months ago

Closed 3 months ago

#5724 closed (fixed)

High cpu usage and low throughput when using BitTorrent DIR-859 A3 / Sudden WAN and LAN router communication loss on multiple builds

Reported by: Sean McCulloch Owned by:
Keywords: Cc:

Description

Router: DIR-859 A3 Build: v3.0-r31277

When downloading via BitTorrent?, throughput is low and CPU usage spikes, causing the router to crash (LAN interface can't be pinged).

Multiple "ath10k_pci 0000:00:00.0: SWBA overrun on vdev 0, skipped old beacon" are logged when the router is slowing down / about to crash.

Attached is the last block from syslog before the crash as well as the subsequent reboot.

Attachments (4)

DIR-859 A3 Crash.txt (37.8 KB) - added by Sean McCulloch 4 months ago.
DIR-859 Top BitTorrent Crash.jpg (142.4 KB) - added by Sean McCulloch 4 months ago.
DIR-859 Top Direct Download Full Speed.jpg (142.3 KB) - added by Sean McCulloch 4 months ago.
DIR-859 Top Direct Download Crash.jpg (139.5 KB) - added by Sean McCulloch 3 months ago.

Download all attachments as: .zip

Change History (16)

Changed 4 months ago by Sean McCulloch

Attachment: DIR-859 A3 Crash.txt added

comment:1 Changed 4 months ago by tatsuya46

check memory usage when its about to crash, and results of "top" in ssh/telnet, im assuming it should be sirq thats maxing the cpu.. and "maximum ports" under ip filter settings are set to what?

Last edited 4 months ago by tatsuya46 (previous) (diff)

Changed 4 months ago by Sean McCulloch

Changed 4 months ago by Sean McCulloch

comment:2 Changed 4 months ago by Sean McCulloch

The first screen grab I uploaded shows the status in top when the router crashes during a BitTorrent? download. "sirq" will increase or decrease depending on the throughput of the download in near realtime. When the router slows down, it's "sys" that spikes (sometimes over 90%). Several processes (ie, hostapd, resetbutton, watchdog etc...) will consume a high percentage of CPU cycles when the router either bogs down or crashes. Router memory consumption increases and the free memory reported in top can be quite log (ie, less than 5MB). While doing a continuous ping of the LAN address of the router I also see waves of high latency (ie over 1000ms) and dropped packets when the router is in this busy state.

The second screen grab shows the direct download of a large .iso file. "sirq" stays in the 22-25% range while maxing out my 50Mb connection. Load average stays very low (the 5 and 15 minutes load averages are higher because the router had just been rebooted after the BitTorrent? download crash). I can do direct downloads at full speed for long periods of time (ie, a full 40GB game download) without any issues.

IP Filter Settings TCP Congestion Control: westwood Maximum Ports: 4096 TCP Timeout: 3600 UDP Timeout: 120

Last edited 4 months ago by Sean McCulloch (previous) (diff)

comment:3 Changed 4 months ago by Sean McCulloch

I did more testing today and have found that disabling the 5Ghz radio seems to fix the high CPU spikes and high memory usage. Any idea why?

comment:4 Changed 4 months ago by Sean McCulloch

I have done further testing concerning this issue on build r31544 and the issue persists when the 5Ghz radio is enabled.

Enabling the 5Ghz radio (QCA9880) consumes around 15MB of RAM on its own (Router has 64MB RAM total and about 61.6MB usable). I'm wondering if this may not be leaving enough free memory for the router to keep running under heavy load. With both the 2.4 and 5GHz radios enabled, I've seen free memory drop to around 5MB before the router crashes.

If I disable the 5GHz radio, free memory stays in the 26-30 MB range and the router runs fine.

Is there a potential fix for the amount of memory the QCA9880 chip consumes (or is there a memory leak?), or is this router simply lacking enough RAM to run this radio properly?

Last edited 4 months ago by Sean McCulloch (previous) (diff)

Changed 3 months ago by Sean McCulloch

comment:5 Changed 3 months ago by Sean McCulloch

The "DIR-859 Top Direct Download Crash.jpg" screen capture shows the status of the router when it crashed while doing a direct download at full speed (~65Mb). Prior to this last Top screen refresh sirq had been much lower (~20-25%).

When it crashes, the following is observed: Wifi and Lan lights on router still blinking, no internet connection (wired or wireless) and can't ping router LAN IP. Wireless clients are still connected to the router (2.4Ghz, the 5Ghz radio is disabled). Reboot the router and it's ok for a while until it is placed under sudden, heavy load.

I can't seem to find a build that doesn't have this issue.

Last edited 3 months ago by Sean McCulloch (previous) (diff)

comment:6 Changed 3 months ago by Sean McCulloch

Summary: High cpu usage and low throughput when using BitTorrent DIR-859 A3 r31277High cpu usage and low throughput when using BitTorrent DIR-859 A3 r31277 / Sudden WAN and LAN router communication loss

comment:7 Changed 3 months ago by Sean McCulloch

Resolution: fixed
Status: newclosed

comment:8 Changed 3 months ago by Sean McCulloch

Resolution: fixed
Status: closedreopened

Any ideas?

comment:9 Changed 3 months ago by Sean McCulloch

Summary: High cpu usage and low throughput when using BitTorrent DIR-859 A3 r31277 / Sudden WAN and LAN router communication lossHigh cpu usage and low throughput when using BitTorrent DIR-859 A3 / Sudden WAN and LAN router communication loss on multiple builds

comment:10 Changed 3 months ago by Sean McCulloch

Have observed the issue on builds 30342, 31277, 31544, 31571

comment:11 Changed 3 months ago by Sean McCulloch

Issue may have been due to having the router use PPOE instead of a static IP for the WAN connection. Uptime of 20 days with 31544 since setting the WAN IP of the router to a static IP in the ISP modem's subnet. Have just flashed 31791, ran some tests and router appears stable.

comment:12 Changed 3 months ago by Sean McCulloch

Resolution: fixed
Status: reopenedclosed

Issue appears fixed by setting router WAN IP to static in the ISP modem's subnet vs. using PPPOE.

Note: See TracTickets for help on using tickets.