configuration about right?

I am constrained (crippled?) by some management-level decisions here,
so please don't think that I selected this topology.

I have two (sometimes more) devices that are required to act as both
directors and real-servers on an intranet.  There is NO back-channel
communication available (don't ask), so all inter-director and inter-
server traffic is through one or two managed switches between the
servers and the clients.  I am to provide virtual service interfaces
(initially, I expect the list to grow) to http servers that front-end
a redundant mySql database, tftp service, and, on some implementations,
streaming media.  The database servers will also be handling other
services for the clients directly.  The linux kernel on the director/
realserver devices is based on 2.6.15.7, with a few drivers added/
changed, plus the -ja patch set (combined patch), on x86 hardware.

In this topology, it seems that ICMP redirects are used for real-server
assignment by the ip_vs code, and the http clients that I have tried
appear OK with this.  I have learned that the tftp clients are not
tolerant of having the data and ack's come from the "other" (non-
director) realserver.  Also, the ip_vs code does not provide connection
tracking (RTSP/RTP combination for streaming) for this topology, so I'm
going to have to create a separate mechanism for that, probably using
nfnetlink.

Here's what I think I CAN do:

I can use the managed switches, which have DHCP/DNS-determined addresses,
as "ping" peers to determine if network connectivity is lost and move the
directors between the devices, as needed, using heartbeat.  I will leave
auto_failback off, just to minimize the disruption of active connections.

I plan to move the http and tftp service, including the associated IP
addresses, with the director.  tftp doesn't work for me through ip_vs,
anyway, so it needs to follow the director, and the value of balancing
the http load is obliterated by the mySql redundancy traffic through the
same network connections (there is a high proportion of updates/commits).

The streaming server will be load-balanced, since nearly all of that is
output data from the realservers, and I expect that load to be very high
when deployed.  I plan to use the wrr algorithm and bias the realservers
away from the active director.

Have I got the layering of directors/realservers about right, or have I
missed something important?

If, after this is deployed, I determine that there is enough tftp
activity to impact the http service (or when they add the next service),
I would like to set up the directors to put those services on separate
realservers, running them on the same realserver only if one of the
realservers is offline.  I haven't figured out how to set that up yet.

One other thing I have not been able to find is the ability to use STP
(spanning tree) packets, rather than adding additional "pings", as any
part of the active knowledge base for heartbeat.  Have I not looked in
the correct places, or is it currently not there?

Thank you.


Disclaimer: The information contained in this transmission, including any 
attachments, may contain confidential information of Panasonic Avionics
Corporation.  This transmission is intended only for the use of the 
addressee(s) listed above.  Unauthorized review, dissemination or other use 
of the information contained in this transmission is strictly prohibited. 
If you have received this transmission in error or have reason to believe 
you are not authorized to receive it, please notify the sender by return 
email and promptly delete the transmission.




-- 
Ultra Monkey - http://www.ultramonkey.org/
To UNSUBSCRIBE, email to [email protected], with a body:
unsubscribe ultramonkey-users [email protected]
where "[email protected]" is YOUR email address.