Cookie based Load Balancing

If 3 Real servers in a non-load balancing environmet are setting session cookies with diffrenet cookie names e.g.
server1 response
set-Cookie: SESSIDSAAAAAA=DMNNNELCECNCKDIIDCPOIMGG
Server2 response
set-Cookie: SESSIDSBBBBBB=DAAMMNELCECNCKPYTWPOIPOP
Server3 response
set-Cookie: SESSIDSCCCCCC=POHYTUOIPOPPLKJHTERIQOKJ
then how can CSM be configured with cookie based stickiness.
I tried cookie insert on CSM with NULL value Assigned to "COOKIE_INSERT_EXPIRATION_DATE".
It resulted in two set cookie responses (one from server and one from CSM).
I am wondering how csm will react ( cookie insert is used) if client request carries two cookie name-value pairs.
clients are behind megaproxy so cookie based stickiness is needed.
Thanks

if you look into a http client request you will see that many times there are more than 1 cookies.
The most important is to make sure the CSM insert a cookie with a different name.
Create your own name.
The client will receive both the csm cookie and the server cookie and will send both when opening a new connection.
The CSM is able to locate its own cookie in the list and do the stickyness.
Gilles.

Similar Messages

Session based load balance + Prepared statements

Experts,
From the docs I understand that there are 3 load balancing techniques. One is client side and two are server side. Of the two, one is session count based load balancing, and as per docs, it is recommended for connection pool setting.
My question is if I have prepared statements originally created using connection to node1, and say if listener re-directs the conneciton to another node node2, will the prepared statement work on node2 ?.
Thanks
Vissu

Just to clarify, the question is:
Are the prepared statements usable when we use session count based load balancing.

Health based load balancing.

I know that RM can provide health based load balancing e.g. RM will stop giving the load if WEF server is not healthy. We have a F5 load balancer, Can't we get the health based load balancing using F5?
Regards Restless Spirit

i think you can do. You can specify the number of monitors that must report a pool member as being available before that member is defined as being in an up state.
check this support article will give you different method of loadbalacing
Load Balancing pool
Please remember to mark your question as answered &Vote helpful,if this solves/helps your problem. ****************************************************************************************** Thanks -WS MCITP(SharePoint 2010, 2013) Blog: http://wscheema.com/blog

IP source based Load balancing?

Hi all;
We encounter the following issue:
A load balancer directs requests in a round robin mechanism to several servers. We want the load balancer direct requests based on the source IP addresses, so that the same host would be directed to the same server at each time it reaquests to be connected (reconnection). Is this possible when using CSM module knowing that NAT is implemented?
Regards

Yes this is possible doing
vserver VAPP
virtual 10.1.1.11 tcp 2514
serverfarm SAPP
sticky 90 group 8
idle 5400
persistent rebalance
inservice
sticky 8 netmask 255.255.255.255 address source timeout 90
This should make the session sticky

URL-Based Load Balancing

I'm having a difficult time trying to configure load balancing on my CSM based on the URL entered. Here is my scenerio:
Two web servers (WebA & WebB), load balanced on a CSM. WebA & WebB have 90% the same content, so most traffic can be load balanced between them without a problem. The problem (for me anyway) comes in where WebA has certain web sites that WebB doesn't, and vice versa. So I need to load balance to both for 90% of the traffic, and point traffic to a particular server the other 10% of the time based on the URL entered.
Below is the test config I have so far (that doesn't work correctly), what I am trying for in this example is that any URL that contains /vhosts/ or /programs/ be directed to WebA, and any URL that contains /platform/ or /ssl/ be directed to WebB, and all other traffic be load balanced between the two evenly. (For testing purposes, the servers are being load balanced in "bridge-mode", in production they will be "routed-mode"....I did't want to go through the change controls to change the IP addresses for the test servers!).
module ContentSwitchingModule 2
vlan 605 client
ip address 10.63.240.4 255.255.255.0
gateway 10.63.240.1
vlan 606 server
ip address 10.63.240.4 255.255.255.0
natpool URL-POLICY-TEST 10.63.240.204 10.63.240.204 netmask 255.255.255.254
map SRV-A url
match protocol http url /vhosts/*
match protocol http url /programs/*
map SRV-B url
match protocol http url /platform/*
match protocol http url /ssl/*
serverfarm URL-POLICY-TEST
nat server
nat client URL-POLICY-TEST
real 10.40.109.100
inservice
real 10.40.109.101
inservice
serverfarm URL-TESTA
nat server
nat client URL-POLICY-TEST
real 10.40.109.100
inservice
serverfarm URL-TESTB
nat server
nat client URL-POLICY-TEST
real 10.40.109.101
inservice
policy TESTWEB-A
url-map SRV-A
serverfarm URL-TESTA
policy TESTWEB-B
url-map SRV-B
serverfarm URL-TESTB
vserver URL-POLICY_TEST
virtual 10.63.240.10 tcp 0
vlan 605
serverfarm URL-POLICY-TEST
sticky 1
persistent rebalance
slb-policy TESTWEB-A
slb-policy TESTWEB-B
inservice

Thanks for the reply Gilles....I've been out of the office for a while.
Well, right now nothing is working....except that all traffic is going to the default server farm assinged to the vserver. Here are the URLs I am testing with:
**************TEST A************
http://10.63.240.10/manual/vhosts/fd-limits.xml
http://10.63.240.10/manual/programs/apachectl.xml
**************TEST B************
http://10.63.240.10/manual/platform/ebcdic.xml
http://10.63.240.10/manual/ssl/ssl_compat.xml
***************BOTH****************
http://10.63.240.10/manual/howto/htaccess.xml
http://10.63.240.10/manual/howto/cgi.xml
When I try attaching to the first URL for example, here is the connection info (I trimmed it down so it will fit here):
MOSL1S1A#sh mod csm 2 real
real server farm Conns/hits
10.40.109.100 URL-POLICY-TEST 1
10.40.109.101 URL-POLICY-TEST 0
10.40.109.100 URL-TESTA 0
10.40.109.101 URL-TESTB 0
MOSL1S1A#
MOSL1S1A#sh mod csm 2 conn
prot vlan source destination
In TCP 605 10.47.10.10:3738 10.63.240.10:80
Out TCP 605 10.40.109.101:80 10.63.240.204:8820
I've tried changing the syntax on the URL statement in the map as such:
/manual/*
*/manual/*
/manual/
*manual*
/manual*

Rv042 dual-wan threshold based load balance?

I have an RV042 (it's old, silver/dark grey plastic front one) w/ firmware 1.3.13.02-tm.
The reason we bought this (long ago) was to balance two WAN connections, one with unlimited data and one capped monthly. It did that once, but for a couple years both connections have been unmetered so it's just been balancing them 50/50. As of today one WAN connection (the new much faster one) is back to being metered but I can't figure out how to configure the RV042 as it once was to prefer sending traffic over the slow, unmetered connection first, and only use the faster metered connection when necessary.
It's been a long time and honestly I only vaguely remember the ability to prioritize a connection based on % of bandwidth used so that all traffic would go over the unlimited connection 1st until it was flooded, and only then fall over to the metered connection. This is totally different than the weighted round robin, or smart link backup.
I found this 3rdparty pforum post that supports that vauge memory and suggests this was eliminated netweem firmware 1.23 and 1.3:
http://www.linksysinfo.org/index.php?threads/rv042-load-balancing-options-from-the-manual-where-to-find.15512/#post-69948
So I humlbly ask... Is it possible to replicate this functionality with the current firmware? if so how? If not, how to do roll back to firmware 1.23?
It sounded like perhaps I could assigned WAN1 a bandwidth of 100000 (even though it's really 1500) and then assign WAN2 a bandwidth of 1 (even though it's really 20000) and the result might be the prioritization I'm looking to achieve... but I feel like I'm stumbling in the dark at the point.
Just FYI, I'm not at all opposed to buying new hardware to acheive this if it's not terribly expensive (ie. <$200). I'd rather not, but I've got to solve this quick.

Hi Jon,
I Also have one of these routers.
On the bottom mine says (v02) which means its hardware version is 2.
I just got this one brand new for home as I have been using them for a very long time now. However I have been using them for VPN and now I am needing the same functionality as you.
I am currently running Firmware Version: 1.3.12.19-tm
If you login to the web management (eg 192.168.1.1) and go to System Management > Dual-WAN
Down the bottom you will see "Protocol Binding".
This is all I know of to send specific ports or applications via a specific WAN.
I'll give you an example of how I am using it currently.. (BTW it seems to be working OK, But you are on a higher firmware)
eg: WAN1 is more reliable than WAN2 which is a cheap unlimited service.
So I bind port 5060 (sip), port 80 (http) and port 443 (https) to WAN1 so that my VOIP phone is on the good service and so is all web traffic.
so all the other stuff can use the unlimited connection.
Also, My current bandwidth settings are
WAN          UPSTREAM          DOWNSTREAM
1                384                       8000
2                384                       10000
And Under: System Management > Bandwidth Management you can also prioritize those ports.
This may help you in some way, So maybe you can help me..
Your post has made me not want to upgrade the firmware.. Can you please confirm that this functionality exists still?
Thanks

Two gateways, port-based load balancing

Hello,
I have a simple question on Mac OS X Leopard/SL Server regarding the use of 2 distinct internet connections on a single LAN.
Gateway #1 : 10.0.1.1 (delivering IPs) - 18 mbps
Gateway #2 : 10.0.1.254 - 4 mbps
Any computer accessing the network is delivered an IP by the DHCP server (10.0.1.1), thus uses #1 as of main gateway.
The main server (10.0.1.16) is running DNS services and a Squid proxy-cache.
Now, is it possible to set all the computers that connect to the network up so that they use the main server as of main gateway and see their requests redirected to #1 or #2 according to the port in use ?
For example:
mail,http,https,jabber -> #1
skype,rtsp,... -> #2
Thank you very much for your help
Tha
Message was edited by: Kwintin

is it possible to set all the computers that connect to the network up so that they use the main server as of main gateway and see their requests redirected to #1 or #2 according to the port in use ?
No. routing is based on destination IP address, not port.
Therefore each client will send all traffic for a specific address to a specific router address. It doesn't matter whether it's talking HTTP, SMTP, IMAP, POP, AIM, or any other protocol - any traffic for that IP will go to the same router.
You have three ways of getting around this.
One is to install a router that supports dual WAN connections. Point all internal clients to the LAN address of the router and let it do the work of routing the traffic as needed, based on its routing policies (routers may be able to route based on port).
Option two is to setup a proxy server for specific services - for example you could setup a HTTP/HTTPS proxy server on a machine that has router #1 as its default gateway and configure the clients to talk to router #2. All traffic on the clients will go over router #2 except the proxied traffic which will go to the proxy and then out via router #1.
This is relatively simple to setup, but is limited to traffic that can be easily proxied (e.g. that probably excludes email).
The third option is static routing. Look at the servers each machine is contacting and setup static routes for the smaller set of addresses. For example, if you're only splitting off traffic to Skype's servers then set each client with a default route of router #1, and static routes to Skype's server to router #2. Now all traffic except that to Skype will use router #1.
This is really only viable if you have a relatively small number of destination addresses you're trying to divert. That's why it works well for Skype (single server address), but wouldn't work well for something more generic such as 'web traffic' since you cannot predict which web servers (and therefore which IP addresses) need static routes.
Of the three options, only option #1 will cover all protocols for all clients, but it's also the only option that costs $$s if your current router doesn't support multiple WAN interfaces.

Sticky load balancing across 2 ports with cookies

Hi,
I have a server configuration where I have 1 top level Apache server that deals with SSL termination (and handles static content) and proxy passes dynamic content onto 2 Tomcat servers on 2 ports, one for http requests (9001) and one for the requests that were secure, but have now been un-encrypted by Apache (9002). My 2 Tomcat servers are load balanced using a CSS and I need this load balancing to stick to the tomcat servers regardless of port so that the user is stuck to the same Tomcat server for their entire session.
I would like to use arrowpoint cookies to perform this stickyness, but the documentation suggests that arrowpoint cookie load balancing (in fact any cookie based load balancing) requires the port to be specified in the content rule. Is this correct? Is my only option to use the source IP for stickyness? I don't understand why the port should be required if the stickyness is via a cookie. Can I not simply configure my 2 tomcat servers as services with no port and add a single content rule that load balances these services using arrowpoint-cookie advanced balancing?
service tomcat1
ip address x.x.x.x
active
service tomcat2
ip address x.x.x.x
active
owner me
   content sticky
     vip address x.x.x.x
     protocol tcp
     url "/*"
     add service tomcat-1
     add service tomcat-2
     advanced-balance arrowpoint-cookie
     active

Angela-
The issue with port is that cookies are very specifically HTTP only and the CSS has no way of knowing what protocol will hit a VIP prior to trying to address it as HTTP. Your issue is actually a bit clearer than it is initially led to be - you can still use 2 different rules by using the configuration below.
However, you might be headed for a headache if you don't implicitly control the client's actions. By default, browsers don't generally send cookies cross-protocol and definitely not cross-domain. Use something like httpwatch or iewatch to check out the headers your client sends to your site. Make sure when the 200ok arrives with the set-cookie that the client sends that cookie in all preceeding packets that are HTTP and HTTPS both.
service tomcat1
string "tomcat1"
ip address x.x.x.x
active
service tomcat2
string "tomcat2"
ip address x.x.x.x
active
owner me
   content sticky9001
     vip address x.x.x.x
     protocol tcp
     url "/*"
     port 9001
     add service tomcat-1
     add service tomcat-2
     advanced-balance arrowpoint-cookie
     active
   content sticky9002
     vip address x.x.x.x
     protocol tcp
     url "/*"
     port 9002
     add service tomcat-1
     add service tomcat-2
     advanced-balance arrowpoint-cookie
     active
With this configuration, the CSS will use the "string" as the cookie value. So if the client were to recieve set-cookie: ArrowpointCookie=tomcat1, it should use it for either rule, and end up on tomcat1 accessing either VIP.
Regards,
Chris

CSS Load Balancing with Cookies

We are trying to load balance 2 backend servers hosted on Websphere with advance balance cookies method.
Restrictions
ServerA is unable to accept cookies generated from ServerB.
ServerA and ServerB are generating random cookies
Unable to modify cookie string with a constant.
How can we load balance based on cookies considering the above restrictions?
We have attempted to do hash based load balancing with cookies but the problem we run into is the servers do not accept cookies generated from another server.
The configuration we tried is written below:
service ServerA
ip address 192.168.10.2
keepalive type tcp
keepalive port 80
active
service ServerB
ip address 192.168.20.2
keepalive type tcp
keepalive port 80
active
content ABC
url "/*"
add service ServerA
string prefix "JSESSIONID="
advanced-balance cookies
port 80
add service ServerB
string skip-length 5
string process-length 16
string operation hash-xor
protocol tcp
vip address 172.16.32.1
active
Can we change the string prefix to JSESSION instead of JSESSIONID= ?
The only place the app guys can add a constant string to match on is before the = sign.
Is it possible for CSS to match on a constant string before = sign e.g below:
service ServerA
ip address 192.168.10.2
keepalive type tcp
keepalive port 80
string id567=
active
service ServerB
ip address 192.168.20.2
keepalive type tcp
keepalive port 80
string id123=
active
content ABC
url "/*"
add service ServerA
string prefix "JSESSION"
advanced-balance cookies
port 80
add service ServerB
string skip-length 0
string process-length 6
protocol tcp
vip address 172.16.32.1
active

It should work.
There is no reason for it not to work...
This is the best method you can have on the CSS for stickyness.
Get a sniffer trace on the client and server with arrowpoint cookie configured on the CSS and capture a failure so we can see what is going on.
also send me the config so I can verify everything is ok.
If you have a service request open with the TAC, you can also give the SR # so I can review what has been done.
Gilles.

Load balancer + Fed auth cookies + Sharepoint

I have a login page which creates a fed auth cookie and establishes session in sharepoint
Now, Since the time site moved to load balancer environment, it started behaving differently
The fed auth cookie is being created. But still I am not logged in!! Which means I am again being redirected to login page for authentication
After another 2-3 attempts, things start working again. Can anyone help as to what is impacting cookies in load balancer? are they being rejected ?

Does the ext directory have the php_oci8.dll? In the original steps the PHP dir is renamed. In the given php.in the extension_dir looks like it has been updated correctly. Since PHP distributes php_oci8.dll by default I reckon there would be a very good chance that the problem was somewhere else. Since this is an old thread I don't think we'll get much value from speculation.
-- cj

Lync 2013 Enterprise load balancing on the front end and edge pool

Hi,
I am setting up a Lync 2013 Enterprise deployment consisting of a Front End pool (x2 FE servers) and an Edge pool (x2 Edge servers). I'm seeing some conflicting advice regarding load balancing using hardware or DNS for the front end and the edge.
On the front end I have 2 internal DNS records 'lyncfepool1.contoso.local' each of which map to one of the IPs of the FE servers. I've used my details to populate the Detailed Design Planner excel spreadsheet and am told that I require a HLB to load
balance my front end pool. I'm aware of the need to load balance HTTPS traffic internally (which will be done by TMG) however other traffic to the front end (SIP, etc) can be balanced by DNS only, and not require a HLB?
Can someone clarify the front end requirement?
Also - looking now at the edge pool - this site again have two edge servers in a pool. We are using a total of six private IP addresses, two per edge service (2 x av.contoso.com, 2 x sip.contoso.com and 2 x webcon.contoso.com). These will be
NAT'ed by the external firewall and directed to the respective external (DMZ) IP addresses on the Edge servers on port 443. I know this isn't true roundrobin due to the intelligence of the Lync client when connecting (in that the Lync client will connect
to one of the public IPs and if it can't connect, it will know to connect to the other service IP), however I want to clarify this set up, particularly the need to direct the external public IP traffic at the DMZ Edge IP specified in the topology builder.
I've attached a basic diagram of the external/DMZ/Edge side which hopefully helps with this question
Persevere, Persevere, Per..

That is because you will always need HLB for a front-end server since it hosts the Lync webservices which use HTTP/HTTPS traffic.
The description on the calculation tool also describes this correctly:
Supports Standard and Enterprise pools (up to 12 nodes), with pure device-based load balancing or a combination of DNS load balancing and device-based load balancing (for
Lync web services)
You can use either Hardware or DNS loadbalancing for SIP traffic only, but you will always need a HLB for the webservices. Both are applicable for the Front-End so you have either
full HLB for both SIP and HTTP(S) traffic
DNS LB for SIP traffic and HLB for HTTP(S) traffic
Hope this is more clear :-)
Lync Server MVP | MCITP Lync Server 2010 | If you think my post is the answer to your question, please mark it as answer so future visitors can easily find it.

Load balancing on iplanet

What are the capabilitiesof the load balancing plug-in "Resonate Command Module" for iplanet? Is this the only load balancing feature that iplanet provides.
Also, if anyone knows of other reasonable (price) load balancing solutions for iplanet web server: preferably, software based only (no load balancing hardware).
I looked at the documentation, and deduced that "resonate" alone cannot provide load balancing for iplanet web servers.
TIA,
Mo

Actually resonate is a sofware based load balancing solution. The confusion is that it doesnt come with the iplanet web server. The command module is only an interface to resonate that will tell it how to react to iplanet internals. So, you would still need to buy Resonate central dispatch for this to be any use to you. Resonate software is expensive. Expect to pay about 60k to start last I checked.
There arent that many other software load balancers that are cheap. I found one open source thing called pen. Check out http://siag.nu/pen
It's really not bad. It supports sticky lb and can actually work for any simple tcp protocol: http,https,ldap,imap,pop,etc....
The only thing is that you need to set up your own failover for pen so that if the pen process goes down another machine will service the requests.

Virtual Hosts for Load Balancing

Hi all,
So I have two identical servers, that have SOA Suite installed pointing to the same database. (RHEL 4) I have configured the mod_oc4j.conf file for round robin, and then I created a virtual host on the httpd.conf file to point to the load balance hardware.
It worked! But with a slight issue. The Load Balance server is orasoaqa.tmpw.net:7777, which opens the main Application Server page. However, when I click on BPEL console, or ESB, it continues using the orasoaqa.tmpw.net:7777 virtual, which of course, breaks even trying to login.
What am I missing here? Here's the entry I added to mod_oc4j.conf:
Oc4jSelectMethod roundrobin:local
Here's the entry I added to the httpd.conf on server 1:
Port 7777
Listen 7777
NameVirtualHost XXX.XXX.XXX.XXX:7777
<VirtualHost orasoaqaapp101.ma.tmpw.net:7777>
ServerName orasoaqa.tmpw.net
ServerAlias orasoaqaapp101.ma.tmpw.net
Port 7777
</VirtualHost>
and on server 2:
Port 7777
Listen 7777
NameVirtualHost XXX.XXX.XXX.XXX:7777
<VirtualHost orasoaqaapp102.ma.tmpw.net:7777>
ServerName orasoaqa.tmpw.net
ServerAlias orasoaqaapp102.ma.tmpw.net
Port 7777
</VirtualHost>
Yes, the port for Application Server and the port on the Load Balancer are exactly the same, I hope this isn't the issue.
Any help would be greatly appreciated.
Thanks!
Message was edited by:
CooperHawkes

Seeing that you have an external load balancer available, you really should not be trying software load balancing. Windows NLBS is a software based load balancing and is not really dependent on the physical/virtual hosting of the OS. There are certain gotchas
w.r.t the NLB modes since they depend on network protocols such as arp, propagation of MAC addresses and usage of the MAC in the L2 routing tables of network devices.
You should be taking the help of the cloud service provider as ONLY they would have an understanding of their underlying networking infrastructure such as VLANS. addressing, routing rules and such.
You could also refer to the VMware support forum for KB's such as
http://kb.vmware.com/selfservice/microsites/search.do?language=en_US&cmd=displayKC&externalId=1006580
Most importantly, NLBS is a base OS related service and you're more likely to get a response in the Windows Server Forum. In the specific case of BizTalk, the Windows NLB is only used for hosting the out of process services such as those associated with
HTTP/s Receive, orchestration published as schemas, web services (asmx) and/or WCF endpoints.
Regards.

Load balancing in Azure PaaS

Hi all,
I'm new to Azure and while our long term goal is to move our entire system into Azure, we have a small feature we'd like to try implementing first so that our infrastructure will be a sort of hybrid model for now. Basically I want to have a WebRole running
a WCF endpoint that reads and writes to a queue. One caller external to my system will call the API to drop off work. My existing system using another method in the API to query for this work that's been dropped off. Pretty simple.
My question is about load balancing. If I were using an F5 or some other hardware based load balancer I'd throw that in front of the Web Roles setup in a round robin configuration and be done with it. if I had 4 instances of the web role and my services
were polling once a second (assuming I have few instances of the polling service running in my existing infrastructure) I'd pick the work up in a second or two. This is perfectly acceptable performance for this process. However, my understanding
is Azure load balancing uses a hash algorithm that works like IP Affinity in a standard load balancer.... this won't work for obvious reasons.... what pattern should I be following for this work? Also note, I'd like to have this geo-distributed as well.
Thanks,
Matt

Hi,
Perhaps, you could use Azure Traffic Manager.
Azure Traffic Manager allows you to control the distribution of user traffic to your specified endpoints,which can include Azure cloud services, websites, and other endpoints. Traffic Manager works by applying an intelligent policy engine to Domain Name
System (DNS) queries for the domain names of your Internet resources. Your Azure cloud services or websites can be running in different datacenters across the world.
More information :
http://msdn.microsoft.com/en-us/library/azure/hh744833.aspx
http://msdn.microsoft.com/en-us/library/azure/dn339010.aspx
http://azure.microsoft.com/en-us/documentation/services/traffic-manager/
Hope this suits your needs.
Regards,
Mekh.

Re: Shared SO's and Load Balancing

Thanks, Tom. I understand this issue a lot better now.
Dale
On Wed, 21 Jan 1998 23:28:33 +0000, Tom O'Rourke wrote:
Dale,
You are right in that it many times does not make sense to load balance
services that are multi-threaded.
But, when a TOOL service object is partitioned with a DBSession service
object, the TOOL service acts as if it is single-threaded regardless of if
it is SHARED or not (while the partition is accessing the database). This
means that the entire partition will be blocked (single-threaded) while the
TOOL service object is using the DBSession to access a database.
Forte blocks the partition to protect the integrity of the data being
passed back and forth to the database because the database vendors have yet
to provide a thread safe call interface to the RDBMS. This is changing as
we speak and Forte is in the process of making the appropriate adjustments.
So, this is the case where it makes complete sense to load balance a
service that is not marked as being SHARED and why it can be a tremendous
performance advantage to use Forte load balancing. As we all know, this
architecture (TOOL EVSO partitioned with DBSession UVSO) is one that is
widely used and proven to produce high performing applications.
The point you bring up is a good one and that is often misunderstood. A
load balanced SO will behave as if it is single threaded (or SHARED).
FYI, we have just added a new feature in release 3.F. Performance-based
load balancing. Check it out.
Tom
At 07:38 PM 1/21/98, Dale V. Georg wrote:
On Wed, 21 Jan 1998 17:24:33 -0000, Richard Stobart wrote:
Dale,
If SHARED is true in a Service Object then the Service Object is not
re-entrant (because many clients are sharing its variables and therefore is
not multitasked (Funny logical naming if you ask me)). I got a lot of replies correcting me on what SHARED means! I do know
what SHARED means, I just phrased it backwards in my post. Excuse me
while I smack myself upside the head. :)
All replicates of a
load balanced partition are not re-entrant and thus equivalent to SHARED =
true. The advantage of load balancing is that the replicates can be
distributed over machines and thus the load is balanced. What I fail to understand is this: If you have a non-SHARED SO all by
itself in a partition which is not load-balanced, it will be
re-rentrant and multiple users can call it at the same time. But as
soon as you load-balance it, all of a sudden it behaves as if it were
SHARED. Why? I don't understand the technical limitations that impose
this, nor do I understand the advantage. For example, let's say that I
have 50 concurrent users of the SO I described above. It's in a
non-load-balanced partition, so all 50 users can access it at the same
time without any problem. Now let's say my server is a little
stressed, so I decide I want to load-balance my SO and have two
replicates, one on the original server and one on a second server. But
now that I've load-balanced it, the partitions act as if they're SHARED
and my 50 concurrent users are going to be lining up in queues and
suffering from horrible response times. How is this advantageous?
================================================
Dale V. Georg
Systems Analyst
Indus Consultancy Services
[email protected]
================================================
================================================
Dale V. Georg
Systems Analyst
Indus Consultancy Services
[email protected]
================================================

Thanks, Tom. I understand this issue a lot better now.
Dale
On Wed, 21 Jan 1998 23:28:33 +0000, Tom O'Rourke wrote:
Dale,
You are right in that it many times does not make sense to load balance
services that are multi-threaded.
But, when a TOOL service object is partitioned with a DBSession service
object, the TOOL service acts as if it is single-threaded regardless of if
it is SHARED or not (while the partition is accessing the database). This
means that the entire partition will be blocked (single-threaded) while the
TOOL service object is using the DBSession to access a database.
Forte blocks the partition to protect the integrity of the data being
passed back and forth to the database because the database vendors have yet
to provide a thread safe call interface to the RDBMS. This is changing as
we speak and Forte is in the process of making the appropriate adjustments.
So, this is the case where it makes complete sense to load balance a
service that is not marked as being SHARED and why it can be a tremendous
performance advantage to use Forte load balancing. As we all know, this
architecture (TOOL EVSO partitioned with DBSession UVSO) is one that is
widely used and proven to produce high performing applications.
The point you bring up is a good one and that is often misunderstood. A
load balanced SO will behave as if it is single threaded (or SHARED).
FYI, we have just added a new feature in release 3.F. Performance-based
load balancing. Check it out.
Tom
At 07:38 PM 1/21/98, Dale V. Georg wrote:
On Wed, 21 Jan 1998 17:24:33 -0000, Richard Stobart wrote:
Dale,
If SHARED is true in a Service Object then the Service Object is not
re-entrant (because many clients are sharing its variables and therefore is
not multitasked (Funny logical naming if you ask me)). I got a lot of replies correcting me on what SHARED means! I do know
what SHARED means, I just phrased it backwards in my post. Excuse me
while I smack myself upside the head. :)
All replicates of a
load balanced partition are not re-entrant and thus equivalent to SHARED =
true. The advantage of load balancing is that the replicates can be
distributed over machines and thus the load is balanced. What I fail to understand is this: If you have a non-SHARED SO all by
itself in a partition which is not load-balanced, it will be
re-rentrant and multiple users can call it at the same time. But as
soon as you load-balance it, all of a sudden it behaves as if it were
SHARED. Why? I don't understand the technical limitations that impose
this, nor do I understand the advantage. For example, let's say that I
have 50 concurrent users of the SO I described above. It's in a
non-load-balanced partition, so all 50 users can access it at the same
time without any problem. Now let's say my server is a little
stressed, so I decide I want to load-balance my SO and have two
replicates, one on the original server and one on a second server. But
now that I've load-balanced it, the partitions act as if they're SHARED
and my 50 concurrent users are going to be lining up in queues and
suffering from horrible response times. How is this advantageous?
================================================
Dale V. Georg
Systems Analyst
Indus Consultancy Services
[email protected]
================================================
================================================
Dale V. Georg
Systems Analyst
Indus Consultancy Services
[email protected]
================================================

Cookie based Load Balancing

Similar Messages

Maybe you are looking for