RE: [Xen-devel] Data broken during FTP test

["Ian Pratt" <Ian.Pratt@xxxxxxxxxxxx>]

> >> However, this problem doesn't happen because of the change in the
> setting.
> >> 1. SG(scatter/gather) flag off
> >> 2. 'mem=1G' is set as a boot parameter, and SG(scatter/gather) flag
> on.

Is this is a 32b install of RHEL5 GA, or x86_64? I presume 32b.

Can you reproduce if you uninstall the RHEL5 xen/kernel and xen tools
and libs RPMs and then install binaries built from xen-unstable?

Rather than disabling SG, what happens if you disable TSO? What about
checksum offload (TX and RX)?

What happens with mem=4G?

This is a worrying bug -- thanks for your help in tracking it down.

Thanks,
Ian



> > This might indicate a high-memory bug in the net driver(s). The TCP
> re-send
> > theory makes no sense -- the decision to re-send is always made by
the
> TCP
> > protocol stack in the domU. dom0 itself will never decide to
re-send.
> >
> > How much memory does your test system have? Are you running 64-bit
> Xen, PAE
> > Xen, or non-PAE 32-bit Xen?
> 
> The environment of the machine is as follows.
> 
> Machine : PRIMERGY model RX200S3
> CPU     : Xeon 5160 3.0 GHz * 2
> Memory  : 12 GB(2048 MB * 6)
> HDD     : AL-9LX 73GB * 2 (Onboard IME RAID1)
> NIC     : Broadcom Corporation NetXtreme BC5715 Gigabit Ethernet (rev
> a3) * 2
>           Intel Corporation 82545GM Gigabit Ethernet Controller (rev
04)
> * 1
> OS      : Red Hat Enterprise Linux 5 GA
> 
> It is confirmed that this problem happens similarly on NIC of the
Intel
> Corporation.
> 
> Best Regards,
> 
> --
> Takanori Kasai
> 


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel