Re: virtio net regression

[Avi Kivity <avi@redhat.com>]

Antoine Martin wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA512
>
> Wireshark was showing a huge amount of invalid packets (wrong checksum)
> - - that was the cause of the slowdown.
> Simply rebooting the host into 2.6.28.9 fixed *everything*, regardless
> of whether the guests use virtio or ne2k_pci/etc.
> The guests are still running 2.6.29.1, but I am not likely to try that
> release again on the host anytime soon! Ouch!
>   


Strange, no significant tun changes between .28 and .29.

Rusty, any idea?

> Antoine
>
> Antoine Martin wrote:
>   
>> Hi,
>>
>> I've got some hosts that were happily running the 2.6.25.x host kernel,
>> kvm-84, kernel.org kvm modules.
>> The guests were running 2.6.25 to 2.6.29.x quite happily.
>> Network was using virtio.
>> Since I upgraded one of the hosts (Intel dual core) to 2.6.29.x
>> yesterday, the virtio network performance of the guests on it dropped
>> dramatically. (for some reason another AMD host did not seem to be
>> affected...)
>> Here are the tests I performed using wget and scp:
>> * guest to guest: fast
>> * guest to host: fast
>> * host to internet: fast
>> * guest to internet: slow!!!
>> I was normally getting ~5MB/s to the host (speed to the internet was
>> limited by the capacity of the DSL line), but since the upgrade the
>> performance had dropped to around 20KB/s!
>> Strangely enough, I could open many new connections to the guest and get
>> more chunks all at 20KB/s!
>> I switched the guests to using ne2k_pci and the performance has been
>> restored...
>>
>> And this is where it gets even weirder...
>> UDP packets get corrupted using ne2k_pci and rtl8139cp but not with
>> virtio...
>> So I can get performance or UDP, but not both...
>>
>> Let me know if there is anything more I can provide to help fix this
>> regression.
>> I can reproduce the problem quite easily without causing problems on the
>> host.
>>
>> Cheers
>> Antoine
>> --
>> To unsubscribe from this list: send the line "unsubscribe kvm" in
>> the body of a message to majordomo@vger.kernel.org
>> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>>     
>
> -----BEGIN PGP SIGNATURE-----
> Version: GnuPG v1.4.9 (GNU/Linux)
> Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org
>
> iEUEAREKAAYFAknmb/sACgkQGK2zHPGK1ruzgwCWPMvAJzToIMbrE7k2K2FHBQlk
> dQCcCpDrTufqIN4ZSQs/dMLTQMYtTAU=
> =lDW9
> -----END PGP SIGNATURE-----
> --
> To unsubscribe from this list: send the line "unsubscribe kvm" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>   


-- 
Do not meddle in the internals of kernels, for they are subtle and quick to panic.