Skip site navigation (1)Skip section navigation (2)
Date:      Wed, 27 Feb 2008 10:06:56 -0300
From:      JoaoBR <joao@matik.com.br>
To:        freebsd-stable@freebsd.org
Cc:        Holger Kipp <hk@alogis.com>
Subject:   Re: em very slow, shared irq... on 6.3p8
Message-ID:  <200802271006.56859.joao@matik.com.br>
In-Reply-To: <20080227104942.GA14530@intserv.int1.b.intern>
References:  <20080227104942.GA14530@intserv.int1.b.intern>

next in thread | previous in thread | raw e-mail | index | archive | help
On Wednesday 27 February 2008 07:49:42 Holger Kipp wrote:
> Hello,
>
> I updated a system with 12 dc-interfaces to a new hardware
> with 14 em-interfaces. Yes, it is a firewall.
> New System is 6.2-RELEASE-p8.
>
> What I now experience between two internal networks (100MBit/s each)
> is the following:
> 1318 packets transmitted, 1317 packets received, 0% packet loss
> round-trip min/avg/max/stddev =3D 0.387/246.153/2441.392/324.142 ms
>
> tcpdump on the firewall shows similar delays (on the outgoing
> interface).
>
> tcpdump on the system I ping however shows very quick responses
> for incoming packages (ie usually less than a millisecond).
>
> I therefore assume that the problem is between receiving the
> irq from em<x> and getting the data from the interface on the firewall
> itself.
>
> My first option would be to activate polling on em-interfaces - but as
> I did not experience this sort of notieceable slowdown with the old
> dc-based firewall (without polling), maybe someone can shed some light
> on this strange behaviour or has other suggestions as well?



I had a setup with 4 (dlink 4port + 1 nic nfe onboard) which run extremely=
=20
stable 6.3

I upgraded the hardware (S939 -> AM2) and used two em 2port cards, without=
=20
polling it was certanly unusable but with polling I got very good performan=
ce

without polling normally in a day or less the machine hung, no msg, simply=
=20
freezed

with polling it stands some days up to two weeks when it freeze again

I upgraded to 7.0 same result

check your setup with vmstat -i and if you see two nics on the same interru=
pt=20
I guess you get the same result as I got

actual I am running one em 2port and two single port pci cards what seems t=
o=20
be stable

anyway, similare setup on Tyan MBs do not have this problem where I also ha=
ve=20
8 nics on one system

I also do not have this problems on S939 boards only on AM2, so I am not su=
re=20
where the problem is exactly but seems in a certain way hardware related

You could try changing your PS unit because you might be short on power wit=
h=20
lots of em cards and SATA disks




=2D-=20

Jo=E3o







A mensagem foi scaneada pelo sistema de e-mail e pode ser considerada segura.
Service fornecido pelo Datacenter Matik  https://datacenter.matik.com.br



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200802271006.56859.joao>