[lug] [O/T]Web Server Fault Tolerance Devices

[Nate Duehr]

Alan, agreed 100%.

For purposes of folks in this forum I would always recommend the LVS
solution over the hardware devices for any site less than 100Mb/sec.

The problem in today's marketplace is that there are an awful lot of
folks who can't find talented people with experience in large network
design... and more importantly those that have the right attitude to run
production networks with the resulting higher bar for quality, uptime,
and general organization skills needed.

Many times, the decision to go with a hardware load-balancer is made by
the human resources they have on-hand at the time.  If they have a
couple of "underworked" Cisco router guys who just finished a big WAN
installation or something, they're probably better off using the Cisco
LocalDirector.  If they have some router-head geeks but no server guys,
Alteons or whatever other hardware box tickles their fancy could be used
with the least pain.  If they have a bunch of Unix-heads, they'll
probably end up doing LVS or a vendor-proprietary or commercial 
implementation of load-balancing on their commercial Unix servers, or 
they'll buy the F5.

None of this observation is scientific by any means, but there's no
clear performance/price winner in the market for load-balancing or
caching yet.  It's still too close to call.  

My employer put in shared Alteons (just so everyone knows my bias), but
if I were a customer I'd never use them... I'd want to be able to
tinker, and you can't tinker with a service from another company the way
I'd want to.  I'd get LVS working.

So in short, yeah... try out LVS!!!  :)

On Tue, Sep 19, 2000 at 08:55:06AM -0600, Alan Robertson wrote:
> Nate Duehr wrote:
> > 
> > Alteon
> > Radware
> > Cisco (LocalDirector)
> > Arrowpoint
> > BigIP/F5 Systems
> > 
> > ...etc etc etc... there are others...
> > 
> > If you're dead-set on a hardware solution (good idea I think), the
> > Alteon's and Cisco's are top-notch, and the price tags match.  Alteon
> > can handle Gig Ethernet without too much trouble.  F5 claims to, but I'm
> > still skeptical of a PC architecture doing that on an overclocked PCI
> > bus.  I have customers that use them, so I guess we'll find out.  Their
> > early machines claimed 100Mb/s performance and couldn't really do more
> > than 70 or so.  Some of that can be blamed on Ethernet and some on the
> > box itself, but Alteons and LocalDirectors could push faster.
> 
> 
> All these folks mainly use PCs in one form or another.  LVS has been tested
> on modest hardware to run a 100mbit link full blast.  I've actually heard
> horrible performance stories concerning LocalDirector.  I don't personally
> have any data on this score though.
> 
> If you buy a pair of gigahertz PCs, and put top-notch NICs on them, I'd put
> LVS (direct routing) up with any of them.  LVS and heartbeat are what
> UltraMonkey is based on.
> 
> Linux is actually quite fast handling ethernet links and routing.  As
> amusing anecdotal evidence, I have a 486/33 with crappy NICs running my NAT
> and firewall at home, and it runs about 10% busy for a .5 mbit link.  If one
> were to take the giant leap to extrapoloating this to a gigahertz athlon or
> so, one can see it's probably in the right ball park for handling gigabit
> networking.
> 
> Many people would find running a 100mbit link more than sufficient in
> practice.
> 
> 	-- Alan Robertson
> 	   alanr at suse.com
> 
> 	-- Alan Robertson
> 	   alanr at suse.com
> 
> _______________________________________________
> Web Page:  http://lug.boulder.co.us
> Mailing List: http://lists.lug.boulder.co.us/mailman/listinfo/lug

-- 
Nate Duehr <nate at natetech.com>

GPG Key fingerprint = DCAF 2B9D CC9B 96FA 7A6D AAF4 2D61 77C5 7ECE C1D2
Public Key available upon request, or at wwwkeys.pgp.net and others.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 232 bytes
Desc: not available
URL: <http://lists.lug.boulder.co.us/pipermail/lug/attachments/20000919/55399976/attachment.pgp>