LVS
lvs-users
Google
 
Web LinuxVirtualServer.org

Re: User base

To: "LinuxVirtualServer.org users mailing list." <lvs-users@xxxxxxxxxxxxxxxxxxxxxx>
Subject: Re: User base
From: Roberto Nibali <ratz@xxxxxxxxxxxx>
Date: Wed, 30 Apr 2003 00:10:08 +0200
Hello,

>We have been using LVS here for just about 2 (?) years now with little to no
issues.  Support here is pretty good - between Joe, Horms, Ratz, Julian, and
Wensong you're going to get dizzy ;)

Before LVS I've used F5s and local directors.  Local directors always seemed
to cause issues with spanning-tree & bridging on our switches.  F5's were
solid but not as flexible.  Obviously price is interesting, too.  Free =
good.

Flexibility is something I do tend to convince risk takers (managers, stake holders, project leaders, CTOs, whatnot ...) with. Take whatever HW load balancer you can afford and you will never be able to implement end-to-end healthchecks customized for special problems and protocols in the way you'll be able to do with LVS.

[I've seen one approach with dynamically programmable FPGAs on a HW load balancer prototype which was designed for companies like Oracle.com. Unfortunately the load balancer development stopped due to money shortage. But we're talking about a starter price of $1.5M, (this is without state transition synchronisation)]

I've probably deployed about 20 load balancers which serve about 100 e-commerce sites, internal proprietary protocols of banks, VPN clusters and whatnot. We have boxes running with old LVS code from back in the 2.2.9 days and do not intend to upgrade since the code is rock solid (that mentioned box does not have security constraints).

One of the features which is very nice is to have a local apache running on different loopback IPs and ports to provide a last server of resort when healthchecking fails. This can be a 3-tier healthcheck over apache reverse proxy to a redundant DB cluster (service handling) or if you would use the per RS treshold limitation you can do limit on service session count (session limit). Depending if all RIPs of a service are taken out by either the service handling (there is a problem with the service providing which has to be fixed) or the session handling (there is a problem with too many people connecting to your service and you therefore need to redirect them to a local page), you'll have different pages of last resort.

You do get first class support here from people all over the world with sometimes very high expertise in various fields of todays IT business. This is something that costs a lot if you need it from any HW load balancer (re-)seller.

I could go on ... if only I had more time.

Best regards,
Roberto Nibali, ratz
--
echo '[q]sa[ln0=aln256%Pln256/snlbx]sb3135071790101768542287578439snlbxq'|dc

<Prev in Thread] Current Thread [Next in Thread>