Opened 2 years ago

Closed 19 months ago

Last modified 4 months ago

#14435 closed bug (fixed)

sudden drop in speed of /dev/net/virtio under VirtualBox (correlated with "virtio_net: ioctl: unknown message 22dd" in serial log?)

Reported by: taos Owned by: korli
Priority: normal Milestone: R1/beta2
Component: Drivers/Network/virtio Version: R1/Development
Keywords: Cc:
Blocked By: Blocking:
Platform: All

Description (last modified by waddlesplash)

VirtualBox: 5.2.18
Host system: archlinux (8 GB RAM, 4 "cores")
Guest: Haiku 64bit (2 GB RAM, 2 CPUs, NAT with virtio-net), at the moment at hrev52302

Quite often (most of the time, already after booting to the desktop) the speed of the network connection drops suddenly (from ~ 4 MiB/s to below 1 KiB/s).

If this happens, the network speed is so slow that even SoftwareUpdater can't always validate the checksums of the repositories but times out. WebPositive can't show the Haiku homepage. Vision can't connect to freenode.

Around the same time, the following messages start to appear in syslog and serial log:

[33mvirtio_net:[0m ioctl: unknown message 22dd

Followed by hundreds of:

Last message repeated 16 times.

If I'm especially lucky, the speed increases again after a while to up to 6 KiB/s and I can try to clone haikuports from github which then often ends after 30 min with

error: RPC failed; curl 18 transfer closed with outstanding read data remaining
fatal: the remote end hung up unexpectedly
fatal: early EOF
fatal: index-pack failed

Change History (6)

comment:1 by waddlesplash, 2 years ago

Component: - GeneralDrivers/Network/virtio
Description: modified (diff)
Owner: changed from nobody to korli

comment:2 by kallisti5, 22 months ago

I see the same issue with my vm at Vultr. Virtio network seems to have some *SERIOUS* issues.

rtl81xx seems to perform *really* well. For a while I thought there could be a performance issue in our network stack... however after running a bunch of bandwidth tests using and the rtl81xx, I think we just have a few bad-apple NIC drivers which don't perform well.

Our network stack *is* slower compared to Linux on like hardware in pretty much all circumstances, but performance is acceptable for our current number of users.

Last edited 22 months ago by kallisti5 (previous) (diff)

comment:3 by waddlesplash, 21 months ago

Please retest after hrev52537.

comment:4 by kallisti5, 21 months ago

I can confirm speeds are greatly improved after this change running the latest nightly. Great work!

comment:5 by korli, 19 months ago

Resolution: fixed
Status: newclosed

comment:6 by nielx, 4 months ago

Milestone: UnscheduledR1/beta2

Assign tickets with status=closed and resolution=fixed within the R1/beta2 development window to the R1/beta2 Milestone

Note: See TracTickets for help on using tickets.