Local Director Weighted Balancing
Hi,
I have LD with sticky cookie-passive and weighted load balancing over 3 servers.
It seems the weights of the 3 server are not respected.
Here are some displays :
LD_Prod# sh sticky
Virtual Machine(s) Sticky
virtualprod:80:0:tcp 240 cookie-passive tolsessionid
LD_Prod# sh weight
Real Machine(s) Weight Timeout
bfpctol3:80:0:tcp 4 never
bfpctol1:80:0:tcp 2 never
bfpctol2:80:0:tcp 2 never
LD_Prod# sh statistics
Real Machine(s) Bytes Packets Connections
bfpctol3:80:0:tcp 530221148 1673183 46326
bfpctol1:80:0:tcp 651629640 2068285 57703
bfpctol2:80:0:tcp 283775703 918391 26235
Note that bfpctol3 has a weight of 4 (doubles the other two server weights), but the number of connections does not shows this situation.
The three servers had been always up since last clear counters command.
Any idea ?
Thanks in advance.
Keep in mind the weighted predictor makes its decision on current, instantaneous incoming traffic. The more traffic you send it the, better the results. Also, I would use the show real command to see the predictor at work. The show real displays the instantaneous connections to each server. The show statistics is the over all number of connections each has been given.
You might also want to consider using the "loaded" predictor. The loaded predictor assigns a weighted portion of connections to a server before moving to the next.
Link to predictor definition:
http://www.cisco.com/univercd/cc/td/doc/product/iaabu/localdir/ldv42/421guide/42ch05.htm#xtocid856564
I hope this helps!
Gonzalo
Similar Messages
-
Local Director Load Balancing question
Using IOS 3.2.2
I have a local director 416 with 2 real servers. I would like to send all traffic to 1 server and if it fails to switch over to the secondary server. The way I see to do it is to use weighted load balancing and set the weight of the 2nd server to zero. But if the primary server fails what will happen since the secondary server has a weight of zero.
JasonNot being a Cisco employee, I guess I can't define what "proper" is, but with the backup command, you don't need to play with the weights at all.
You just bind the primary real server to the virtual, then assign a backup to the primary. The backup server does not get directly bound to the virtual, so the only way it's going to get any traffic is if the primary server is down. -
Server load balancing on Local Director implementing HTTP redirection
Hi, got an interesting problem when implementing the HTTP redirection method of session persistance or stickiness when balancing a couple of web servers on a Local Director 416. The session persistance works fine but it looks like one of the two servers is getting all the traffic rather than balancing accross the two. The servers are equally spec'd and when it comes to the predictor setting for the virtual, I'm just using the defaults i.e. leastconns/roundrobin. Is there a better predictor setting to use?? Any input would be appreciated.
Rgds
Adam LacobaThe loaded argument has worked well. Take note that this requires a value to be set with the weight command. It gives a specific amount of traffic to one, and then moves to the other.
-
Port based LB and Local Director
My customer would like to know whether the Local Director supports Load Balancing using the source port information. If this is possible any document on the configuration wouldbe highly appreciated.
Regards,
Maheshhttp://www.cisco.com/en/US/products/sw/iworksw/ps2769/products_maintenance_guide_chapter09186a008007d9fa.html#xtocid225795
as per this document i think the loadbalancing based on source port information cannot be done.
''The MNLB Services Manager makes the load-balancing decisions based on application availability, server capacity, and load distribution algorithms such as round robin or least connections, or the Dynamic Feedback Protocol (DFP).'' -
Local Director & HTTP Redirection
Hi, just wondering if anyone has successfully implemented HTTP
redirection on Cisco Local Director 416 running ver 4.1.2 Basically I'm
after any pitfalls to look out for. Current implementation utilizes
generic session stickiness however this has problems for clients
trying to access our load balanced servers if they come from behind
multiple proxies via their ISP.
Servers are Private IP addressed (Firewall in front of LD performing NAT static
txlation)
On another point, how does the Cisco CS11000 (Arrowpoint) perform
session persistance/stickiness or what "better" ways does the
Arrowpoint acheive this as compared to Local Director?
Any helpful input would be much appreciated.
Rgds
AdamSince there has been no response to your post, it appears to be either too complex or too rare an issue for other forum members to assist you. If you don't get a suitable response to your post, you may wish to review our resources at the online Technical Assistance Center (http://www.cisco.com/tac) or speak with a TAC engineer. You can open a TAC case online at http://www.cisco.com/tac/caseopen
If anyone else in the forum has some advice, please reply to this thread.
Thank you for posting. -
On Cisco site Local Director 400 series has been classified as End of Sales. Does anybody know what product replaces them?
Also I would like to know if IOS Server Load Balancing feature does the same job of Local Director as far load balancing between Web server s is concerned.
Thanks for any input.The Local Director has a few potential replacements.
IOS SLB can perform the same function and if you have a 6500 already, it may be a good choice.
Also, the Content Swtiching Module (CSM) is more robust and has additional fuinctionality such as one-armed load balancing and firewall load balancing, among others...
The Content Services Switch (CSS) is a separate appliance that also has additional functions from the LD. It is comparable to the CSM in features.
Both the CSM and CSS have optional SSL modules to integrate SSL temination allowing the Load Balancer to do layer 5 inspection of "encrypted" traffic.
Hope this helps!
-Steve -
Local Director unable to detect a failed web site
Hi all,
I have configured the local director to load balance several web sites on two real servers. One of these sites is required to be configured in backup mode.
For some reason the local director is unable to detect the site has gone down (by stopping the site to update the web content), therefore not directing requests to the backup real server.
I am using IIS 5.0, since its quite popular out there, has anyone experienced a similar situation and if so, how did you resolve it?
thanksYour problem is based in the Operating system, not the Local Director. The LD is a layer 3 device. Microsoft put their application architecture together different from the Unix world, ie: Any ip address bound to a MS box is answered for by the OS - not the application (IIS in this case) If IIS is hosed up the OS will still answer for the IP you have bound to the NIC of the OS - the OS does not care at layer 3 that you want a higher level reply - LD does its availability checking at Layer 3 - the OS layer for MS. In the Unix world when an application is bound to a port - ie: Apache to port 80, then the apache deamon answers for that ip/port - not the OS - Apache panics and it will not respond at Layer 3 - the OS could be fine, but the LD will know that ip/port is not responding and will take the appropriate actions by your LD config.
Check out the httpprobe feature of the LD - it lets you test for a file in a URL inside the LD.
Cheers -
Local Director Remote Management
Is there any way to use SNMP to manage the Local Director. I have the SNMP community but I can't seem to change the server status from out of service to in service using snmpset (from UCD-SNMP). Am I doing this in vain. Are there any ways to do remote management of the LD or can it easily be scripted. Thanks.
you cannot SET values for the Local Director through SNMP. Just GET them.
There used to be a GUI for the LD, that works up to version 3.x. It no longer works in version 4.x. The name for it is CVS and it can take servers out of service. It does it by telnetting to the LD and issueing commands.
Also, the LD support Dynamic Feedback Protocol (DFP) which is a protocol that allows the LD to talk to an agent application (installed on the server) and can dynamically adjust the server weight according to load.
Hope that helps,
-Steve -
Local Director 416 Seamless Real Changes
Say I have a Virtual:443 with two reals web01:6443 and web02:6443. Now I want to add in reals web03:6443 and web046443 and remove web01:6443 and web02:6443. But I don't want to disrupt any conversations that may already be in progress to one of those servers. How do I migrate to the new without disturbing the connections on the old but not allow any new connections to be made to the old? Thanks.
There is a well known problem of keeping a session bound to one physical real server after the session is setup on HTTP (open) transport and later moved to HTTPS (secure) transport protocol. In this case, the HTTP redirection should be setup at the application level. The server can send a redirect (either a 302 or an HREF embedded in the actual web page itself) to HTTP, so no load balancing takes place at the Local Director. Another possible solution would be to switch to HTTPS sooner, and perform the session opening encrypted. This is preferred also for encrypting the login username and password. If you keep the whole session running via HTTPS, the generic sticky can ensure that the same server will be used for a client. The second solution and configuration is described in this document. The HTTP traffic is limited to first contact with the site (the client can type simple site name to address bar). When reaching Local Director, the browser receives a 302 redirect message with the URL configured, including the HTTPS tag, so there is the complete name of the server to reach. The session (on application level) is built from the beginning with the secure server (the real servers do not handle HTTP at all). Tthe session continues with the unique server name, so it is always reaching the same.
http://cisco.com/en/US/products/hw/contnetw/ps1894/products_configuration_example09186a00801a7299.shtml#maintask1 -
Local director 417 replacement? CSS 11501 software 8.20
Hi,
i would like to ask if CSS 11501 (software 8.20) is still the replacement hardware for local director 417? currently we use this local director to load balance http and https traffic using a flat network implementation meaning virtual IP is in the same netwoork as the physical server. will CSS 11501 can support this implementation?
Current Local Director setup example
web server virtual IP 10.10.10.1
real server ip 1 10.10.10.2
real server ip 2 10.10.10.3
rgds,
mc.malinaolocal director is well before my time, so I can't comment on those and their official replacement.
I'd say you're better off going straight to ACE modules or appliances.
They're the latest generation of Cisco's load balancers.
We're very happy w/ our ACE-20 modules. -
CSS and Local Director comparison
I am new to this technology and I want to know what are advantages and disadvantages of Local director over the CSS and vice versa besides the price. What functionality does one has that is not supported by the other ?
If you are looking at implementing a load balancer then I would suggest to go for the CSS. The CSS is the next generation of load balancer and will replace the Local Director. The CSS functionality and extensibility is far greater than that of the local director. Another point is that feature developement is far more aggessive on the CSS.
Phil -
Using the Cisco Local Director
Hello All-
We are attempting to set up a server farm that is load balanced with the Cisco Local Director and are experiencing difficulties with some clients whose networks have multiple proxy servers. We have used both the generic sticky command as well as the cookie-insert sticky without total success. Our next option looks like we should put our own proxy server in front of the LD and use SSL. Has anyone experienced these problems and come up with a solution. Unfortunately we don't have weblogic clustering licenses so our state is lost whe a client gets redirected to a new server. Thanks for your time!
-Adam
[email protected]
415.887.9106In directed mode, the client browsers are configured to point to a proxy to a virtual IP address on the Local Director. I know the Local Director does not support a real server that does not have a virtual address defined so it can't transparently forward like the CSS.
-
Migrating Local Director to CSM - commands "name" "oos"
On a CSM v4.2(3) on a 6509: I'm trying to make the real servers have names (like on the local director) instead of IP addresses:
VOC-SW15#conf t
Enter configuration commands, one per line. End with CNTL/Z.
VOC-SW15(config)#mod csm 4
VOC-SW15(config-module-csm)#serverfarm TEST1
VOC-SW15(config-slb-sfarm)#real ?
A.B.C.D real server IP address
name use a named real server
VOC-SW15(config-slb-sfarm)#real name webtest
% Config not found for module real WEBTEST
VOC-SW15(config-slb-sfarm)#real name ?
WORD real server name
VOC-SW15(config-slb-sfarm)#real name webtest
% Config not found for module real WEBTEST
VOC-SW15(config-slb-sfarm)#
VOC-SW15(config-slb-sfarm)#real webtest
VOC-SW15(config-slb-module-real)#address 10.1.250.37
This is what I end up with (I want the real server under server farm test1:
real WEBTEST
address 10.1.250.37
serverfarm TEST1
nat server
no nat client
real 10.1.250.38
inservice
Also, if anyone has an easy way to take a box out of service (like on the local director) please let me know. On CSM we're having to do this:
VOC-SW15#conf t
VOC-SW15(config)#mod csm 4
VOC-SW15(config-module-csm)#serverfarm relays
VOC-SW15(config-slb-sfarm)#real 10.1.11.23
VOC-SW15(config-slb-real)#no inservice
on ld it was just: oos real relay-01
Thank you for all the posts. They have been tremendous help in getting the CSM in production.
ps (Gilles, My Christmas wish is that you're working on a CSS/CSM Book for Cisco Press. One is definately needed)
Thank you, SheilaWe figured out the name configuration.
real MLXWEBTEST
address 10.1.250.36
inservice
serverfarm TEST1
nat server
no nat client
real 10.1.250.38
inservice
real name MLXWEBTEST
no inservice
real server farm weight state conns/hits
10.1.250.38 TEST1 8 OPERATIONAL 0
MLXWEBTEST TEST1 8 OUTOFSERVICE 0
Is this the best configuration? We have to migrate
250+ servers over to the CSM. Thanks, sheila -
Hi!, I have a question: Is there any way using CISCO Local Director instead of WebLogic Clustering feature ? because that our production folks insist on using "hardware" solution. We us iPlanet Web server + WebLogic Application Server in our application. Very appreciate if anyone can answer my question.
Not yet. First, I do not believe that there is a hardware load balancer
that "understands" WL stickiness, especially WRT failover. Only WL's
plug-ins understand that, so right now you must use them.
I expect that WL will eventually be able to route a request to the "right"
server if it shows up at the "wrong" server, meaning that eventually each WL
instance in a cluster could be used as the front-end of the cluster, so a
hardware solution could work very well for non-session requests (where it
doesn't matter which WL instance gets the requests) and for session requests
(where it must go to the "right" server) it will re-route automatically.
Right now I suggest using NAS to front the cluster; it appears to be the
best solution. If you use a hardware load balancer, then a group of Apache
servers on Linux would be as good or better of a solution to front the
cluster (i.e. load balancer distributes to a set of Apache servers that are
working independently of each other).
Cameron Purdy
[email protected]
http://www.tangosol.com
WebLogic Consulting Available
"Bernie Wong" <[email protected]> wrote in message
news:[email protected]...
> Cameron,
>
> Can I use a hardware load balancer that respects stickiness and therefore
> not use WL-based HTTP balancing.
>
> "Cameron Purdy" <[email protected]> wrote in message
> news:[email protected]...
> > From what I have heard, you can use CISCO local director to spread out
> > incoming HTTP requests across a number of Apache or IIS or NAS servers,
> each
> > of which has the WL cluster feature. That really only makes sense
though
> if
> > you have too large an amount of web server requests (typically static
> files
> > like HTML, GIF, etc.). There is no way currently to remove the WL
cluster
> > feature that plugs into the web server. If the request goes to the
wrong
> WL
> > server, it won't get re-routed to the right WL server. There IS such a
> > thing as a "wrong" server because HTTP sessions are sticky. Following
the
> > must-use-hardware line of thought, though, you could theoretically have
10
> > identically configured boxes running Apache with the WL mod and also
have
> WL
> > on each box, then local director could hardware distribute to the 10
boxes
> > that would then software distribute to the same 10 boxes. In other
words,
> > each host in a 10-host WL cluster is also running Apache with the WL mod
> (or
> > one of the others).
> >
> > --
> >
> > Cameron Purdy
> > [email protected]
> > http://www.tangosol.com
> > WebLogic Consulting Available
> >
> >
> > "Think Lee" <[email protected]> wrote in message
> > news:[email protected]...
> > Hi!, I have a question: Is there any way using CISCO Local Director
> instead
> > of WebLogic Clustering feature ? because that our production folks
insist
> on
> > using "hardware" solution. We us iPlanet Web server + WebLogic
> Application
> > Server in our application. Very appreciate if anyone can answer my
> question.
> >
> >
>
>
-
Local Director SNMP for CPU and Memory
Is there any way to get the CPU and memory utilization from a Local Director using SNMP? When I do an SNMP walk, I don't see any objects for CPU and memory.
Thank youThere's a "Show cpu " commmand availabe. Type help or '?' for a list of available commands.
Maybe you are looking for
-
Is it possible to generate a PDF and send it as a mail attachment ?
Hello All, While there are many contributions and suggestions on how to send a pdf form via email, I will like to find out if its possible to send a pdf as part of an email attachment WITHOUT even viewing it first ? Below is a brief descrption of w
-
It says I have a bad connection or something but on my 1st Ipod Touch I CAN set up imessage and facetime. On my second one I can't. Why is that?
-
Data collection: Conversion exit and Input Conversion of Local Currency
Hi, I'm collecting transaction data from BI where the local currency for one company has no decimal places. Therefore if I collect the data with no conversion, the value is multiplied by 100 in BCS . has anyone used the Conversion Exit or Input Conve
-
Burn failed: there was an error producing data for the burn
Getting the following message after trying to bounce a project to CD: Burn Fail: There was an error producing data for the burn. Using Logic Pro 9. Burn has worked with all projects up til now. Not sure what is different about this one. Still works w
-
How do I unlock an object?
Manage to figure out how to select object and lock it now I cant find how to unlock it, please help?