Mailing List Archive

At 14:36 13/03/2001 -0500, you wrote:

>I'm trying to build a production LVS server.
>
>What is the best combination of kernel (2.2 or 2.4), lvs code? I would
>like to be fairly bleeding edge but this is production. I will have two
>lvs servers with keepalived running. I'm also planning on using
>ldirectord for the service monitoring, is that still the best choice?

Hi Matt,

What kind of service are you planing to loadbalance ? smtp, http, ssl, ...?

Your LVS design depend on your need (ie: bandwidth, ...) if you plan to use
LVS NAT your design is correct. Just think in HA using a hearthbeat or
other stuff (i am working on VRRP/HSRP right now to add a strong routing
level tolerence) => With that kind of stuff you add to your main LVS server
a Backup server and you can even loadbalance flows between the two LVS
server (master & slave).

LVS 2.2 code is really stable, but if you plan using LVS with source NAT
for example you will need 2.4.

As an example, I run in production 3 LVS (using keepalived) loadbalancing 4
servers each over HTTP & SSL protocols. It really works fine

Hope it will help you,

Best regards,

Alexandre

"Matthew S. Crocker" wrote:

> I'm trying to build a production LVS server.
>
> What is the best combination of kernel (2.2 or 2.4), lvs code? I would
> like to be fairly bleeding edge but this is production. I will have two
> lvs servers with keepalived running. I'm also planning on using
> ldirectord for the service monitoring, is that still the best choice?
>
> Thanks
>
> -Matt
>
> --
> ----------------------------------------------------------------------
> Matthew S. Crocker
> Vice President / Internet Division Email: matthew@crocker.com
> Crocker Communications Phone: (413) 587-3350
> PO BOX 710 Fax: (413) 587-3352
> Greenfield, MA 01302-0710 http://www.crocker.com
> ----------------------------------------------------------------------
>
> _______________________________________________
> LinuxVirtualServer.org mailing list - lvs-users@LinuxVirtualServer.org
> Send requests to lvs-users-request@LinuxVirtualServer.org
> or go to http://www.in-addr.de/mailman/listinfo/lvs-users

Go with the ultramonkey solution that is what I us here at work for our web
servers. I t works great. :)



--

Andy Gussie
Network Administrator
eFruit International, Inc.
7380 Sandlake Rd Suite 400
Orlando FL, 32819
Ph: 407-352-8081 Fax: 407-352-8085
mailto:andy.gussie@efruitinternational.com
http://www.efruitinternational.com

"Matthew S. Crocker" wrote:

> > Hi Matt,
> >
> > What kind of service are you planing to loadbalance ? smtp, http, ssl, ...?
>
> HTTP, HTTPS, FTP, IMAP, POP3, SMTP
>
> I plan on using Direct routing
>
> Two LVS Servers
> 4 Worker machines
> 1 Network Appliance Netfiler 720 with the data
>
> Everything connected with a cisco 3548XL switch.
> Each worker machine will have a seperate 10/100 NIC for NFS to the NetApp
> which has a 1000Base-T connection to the switch.
>
> I plan on routing a subnet to the cluster VIP from our router and LVS the
> subnet to the working machines.
>
> Example:
>
> LVS VIP = 204.97.12.5
> Clustered Subnet = 204.213.240.0/27
> Worker machines = 204.97.12.6 - 9
>
> Cisco has this static route.
>
> ip route 204.213.240.0 255.255.255.224 204.97.12.5
>
> The lvs servers are arp'ing for 204.97.12.5 and have either sub-interfaces
> or fwmark settings for the 204.213.240/27 subnet. Working machines are
> also on the 204.97.12.x network with dummy aliased interfaces for the
> subnet. I don't think I need the worker machines listening for
> 204.97.12.5 at all because they won't be getting any direct traffic on
> that. If an lvs server crashes heatbeat should switch to the lvs server
> which can arp for 204.97.12.5 and pickup the subnet from the router.
>
> Does this scheme make sense? Will it work?
>
> Thanks
>
> -Matt
> --
> ----------------------------------------------------------------------
> Matthew S. Crocker
> Vice President / Internet Division Email: matthew@crocker.com
> Crocker Communications Phone: (413) 587-3350
> PO BOX 710 Fax: (413) 587-3352
> Greenfield, MA 01302-0710 http://www.crocker.com
> ----------------------------------------------------------------------
>
> _______________________________________________
> LinuxVirtualServer.org mailing list - lvs-users@LinuxVirtualServer.org
> Send requests to lvs-users-request@LinuxVirtualServer.org
> or go to http://www.in-addr.de/mailman/listinfo/lvs-users

http and https

--

Andy Gussie
Network Administrator
eFruit International, Inc.
7380 Sandlake Rd Suite 400
Orlando FL, 32819
Ph: 407-352-8081 Fax: 407-352-8085
mailto:andy.gussie@efruitinternational.com
http://www.efruitinternational.com

> Hi Matt,
>
> What kind of service are you planing to loadbalance ? smtp, http, ssl, ...?

HTTP, HTTPS, FTP, IMAP, POP3, SMTP

I plan on using Direct routing

Two LVS Servers
4 Worker machines
1 Network Appliance Netfiler 720 with the data

Everything connected with a cisco 3548XL switch.
Each worker machine will have a seperate 10/100 NIC for NFS to the NetApp
which has a 1000Base-T connection to the switch.

I plan on routing a subnet to the cluster VIP from our router and LVS the
subnet to the working machines.

Example:

LVS VIP = 204.97.12.5
Clustered Subnet = 204.213.240.0/27
Worker machines = 204.97.12.6 - 9

Cisco has this static route.

ip route 204.213.240.0 255.255.255.224 204.97.12.5

The lvs servers are arp'ing for 204.97.12.5 and have either sub-interfaces
or fwmark settings for the 204.213.240/27 subnet. Working machines are
also on the 204.97.12.x network with dummy aliased interfaces for the
subnet. I don't think I need the worker machines listening for
204.97.12.5 at all because they won't be getting any direct traffic on
that. If an lvs server crashes heatbeat should switch to the lvs server
which can arp for 204.97.12.5 and pickup the subnet from the router.

Does this scheme make sense? Will it work?

Thanks

-Matt
--
----------------------------------------------------------------------
Matthew S. Crocker
Vice President / Internet Division Email: matthew@crocker.com
Crocker Communications Phone: (413) 587-3350
PO BOX 710 Fax: (413) 587-3352
Greenfield, MA 01302-0710 http://www.crocker.com
----------------------------------------------------------------------

> > What kind of service are you planing to loadbalance ? smtp, http, ssl, ...?
>
>HTTP, HTTPS, FTP, IMAP, POP3, SMTP
>
>I plan on using Direct routing
>
>Two LVS Servers
>4 Worker machines
>1 Network Appliance Netfiler 720 with the data
>
>Everything connected with a cisco 3548XL switch.
>Each worker machine will have a seperate 10/100 NIC for NFS to the NetApp
>which has a 1000Base-T connection to the switch.
>I plan on routing a subnet to the cluster VIP from our router and LVS the
>subnet to the working machines.
>
>Example:
>
>LVS VIP = 204.97.12.5
>Clustered Subnet = 204.213.240.0/27
>Worker machines = 204.97.12.6 - 9
>
>Cisco has this static route.
>
> ip route 204.213.240.0 255.255.255.224 204.97.12.5
>
>The lvs servers are arp'ing for 204.97.12.5 and have either sub-interfaces
>or fwmark settings for the 204.213.240/27 subnet. Working machines are
>also on the 204.97.12.x network with dummy aliased interfaces for the
>subnet. I don't think I need the worker machines listening for
>204.97.12.5 at all because they won't be getting any direct traffic on
>that. If an lvs server crashes heatbeat should switch to the lvs server
>which can arp for 204.97.12.5 and pickup the subnet from the router.
>
>Does this scheme make sense? Will it work?

Each worker machine own 2 NIC, one directly connected to your appliance and
the other to the switch ?

=> or your two NIC interfaces are in the same 204.97.12.0/27 subnet
directly connected onto the switch ?

so you are using 204.97.12.6 204.97.12.7 204.97.12.8 204.97.12.9
204.97.12.10 204.97.12.11 204.97.12.12 204.97.12.13 for your worker
machines. So you use a 1Gb interface to handle simultaneous bandwidth
(agregation) with your workers machine. All the net interface are in the
same network.

If the WAN clients users (for your loadbalanced service : smtp, pop, imap,
http, ssl, ftp) do not use more than 100 MBit/s bandwitdth, the LVS direct
routing is not needed. You can use a simple LVS NAT because all the servers
are on the same network bus and if a server want to talk with another
server on the same bus (netapp) then an icmp redirect will be sent and you
will not bridge directly the traffic with your LVS.

Hopes it will help you,

Regards,

Alexandre