Re: Running haproxy on the cluster nodes

Thanks Willy. I guess it's time to ask about swapping out the NIC or just deal with the performance I'm getting. Unfortunately, 100Mbps is all we have on the Internet side, so this may all be moot anyway -- but I wanted to try to rule out any infrastructure issues on our side, since who knows what we'll want to do later.

Out of curiosity, going back to what I mentioned earlier -- is it considered a bad practice from a bandwidth utilization standpoint to have the client, proxy, and cluster nodes all on the same segment, since (a) it puts double the work on the proxy's single NIC connected to that network, and (b) the data requested by the client has to hit that network twice, as it travels from the cluster node to the proxy and then again from the proxy to the client?

The reason I ask is that it seems that situations like this could come up -- for example, if you have a web server (client) talking to a cluster of database boxes (servers). It seems to me that the issue described above, where the data is hitting the network twice, would occur here too. Would it be considered preferable to have the database servers sit on a separate network, with the proxy machine sitting between the two, in order to lessen the amount of noise on the wire?

Thanks,
Martin

On 12/12/07, Willy Tarreau <w#1wt.eu> wrote:
>
> On Wed, Dec 12, 2007 at 05:32:53PM -0500, Martin Goldman wrote:
> > These are hosted dedicated servers, so I don't have much info on the
> > hardware. But from looking at dmesg, it looks like I may have one Intel
> PCI
> > GigE card (which is connected to the LAN) and and a Broadcom/Tigon3 PCI
> > Express card (which is connected to the Internet):
> >
> > ----------
> > e1000: 0000:08:02.0: e1000_probe: (PCI:33MHz:32-bit) [MAC_address]
> ^^^^^^^^^^^^^^^^^^
> Here is the culprit !
>
> > e1000: eth1: e1000_probe: Intel(R) PRO/1000 Network Connection
> > e1000: eth1: e1000_watchdog: NIC Link is Up 1000 Mbps Full Duplex
> >
> > eth0: Tigon3 [partno(BCM95751) rev 4201 PHY(5750)] (PCI
> > Express)10/100/1000BaseT Ethernet [MAC address]
> > tg3: eth0: Link is up at 100 Mbps, full duplex.
> > ----------
> >
> > I googled the Intel card (model 82541PI according to lspci) and it is
> indeed
> > a PCI 33/66MHz card,
>
> Yes that's it but it's most often connected in 33 MHz as reported above,
> explaining the limitation.
>
> > and the Broadcom (model BCM5751) is indeed a PCI Express card.
>
> It's sad that it's limited to 100 Mbps on the wire, because in this
> case the 300 Mbps limitation will never even be reached.
>
> > Would having one of each type of card like this allow for the
> > type of bus saturation that you describe below?
>
> not due to one of each type, just the fact that the intel one is
> limited here. The PCI express one should not cause trouble though
> but it's currently connected to a 100 Mbps network.
>
> > I'm not sure if the hosting company will do it, but should I
> > look into having the PCI card swapped out for another PCI Express?
>
> At least if they could swap both of them it would be better,
> but if you intend to scale to the gig with 2 NICs, you will
> get the problem again as soon as you turn the internet switch
> to 1 Gbps. Yes, you can try to ask them if they accept to plug
> a second PCIe INC in the server to replace the on-board one,
> which could serve as an admin card only for instance.
>
> > Many thanks for all the explanation; as you can obviously tell, I'm not
> much
> > of a hardware guy :)
>
> No problem :-) The hardware is often the problem when numbers are
> not good and sysctls are OK.
>
> Regards,
> Willy
>
>
>
Received on 2007/12/13 02:42