Load Balancing -- preserve client IP?

Does the Flash Media Streaming Server need to know the client's IP address in a load-balanced environment? 
We will be doing straight-forward hardware load balancing -- we are not architecting the cluster for communication between Flash Media Servers.
We're trying to determine whether we need to configure the hardware load balancer to use SNAT to preserve the original client IP, which for us has ramifications about subnet choices for both the LB and server.
Thanks!

I dont think the server would need the client's IP. How do you plan to server your content -  RTMP or HTTP ?
Shiven

Similar Messages

  • DPS 6.3.1 Load balancing with client affinity?

    I have 4 directory servers. 2 masters and 2 consumers. They are all in the same data source pool.
    I am using proportional load balancing. The masters are weighted for writes, zero for reads. The consumers are weighted for reads, zero for writes.
    Client affinity is on and set to "set at write and use for write operations"
    My goal is to have writes go to the same master. I have achieved that but have created this odd loop affect for reads though.
    In my ldap admin app when I add an attribute to an entry I have to refresh it to see the new attribute. Makes sense according to the config above. How should I configure this to stop this from occurring? And due to my proportional weights is setting client affinity neccesary? Or perhaps I am using the wrong client affinity setting.
    Edited by: Xoth on Aug 18, 2009 3:37 PM
    I wanted to add that the read write state in the data source properties for the directory servers is set to read/write for the masters, and read only for the consumers. So maybe its overkill to set this, in affect again, in the proportional weights for load balancing.

    If your ldap admin app doesn't have a constant update, which would be a resource hog, you will have to refresh to see the changes.
    You don't want the app updating itself from your right operations. The information that is changed in the consumer proves that the write was performed and replication did occur.
    If you are writing to your supplier you should read from your consumer to insure that the update occured through the supplieer to the consumer. If you admin app doesn't automatically update you will have to do it manually.

  • Load balancer + preserving IPs

    I seem to have encountered a rather severe limitation of the mod_loadbalancer, and I'm hoping there is a workaround.
    I have a SJSAS 8.1EE cluster fronted by Apache (also tried SJS WS6.1 too) with SJSAS's loadbalancer plugin.
    It seems that the appserver instances are seeing the IP of the loadbalancer instead of the original client IP address (web browser).
    Several apps we run require to check the IP of the client system (for authentication and other reasons), so this is kind of an issue given mod_loadbalancer is now replacing them with it's IP.
    Previously I've used mod_jk with JBoss clustering, and there's no such problem because they intelligently use the AJP protocol, so IP's are preserved.
    Surely this would be a common scenario for ppl using loadbalancers, so I'm wondering if there's any way around it such that the destination appserver sees the client's IP, even when going through the loadbalancer.
    Another idea that comes to mind is the question of examining another HTTP tag (eg <X-forwarded-by>) to get the client's original IP.
    Any ideas would be greatly appreciated!

    i believe i'm looking at the same issue. We have struts tags in our jsp's that are returning the incorrect client url from the loadbalancer. where i'm expecting a link with the url "https://mysite.com/", it's returning "http://mysite:443".
    our configuration consists of a sun webserver (v6.1) with the lbplugin pointed to a cluster of two sun app servers (v8.1). the loadbalancer.xml is configured with the https-routing param set to false, meaning that ssl connections should be forwarded to the http port of the app servers.
    the same problem (i believe) is documented elsewhere in sun's bug site.
    http://bugs.sun.com/bugdatabase/view_bug.do?bug_id=6269102
    http://bugs.sun.com/bugdatabase/view_bug.do?bug_id=6188932
    http://bugs.sun.com/bugdatabase/view_bug.do?bug_id=4814778

  • H-REAP and Client Load-Balancing

    I'm told by Cisco that H-REAP does not support client load-balancing.
    We have a situation where we want to deploy LWAPPs using H-REAP into a conference room where training would take place.
    Any suggestions on how to overcome the inevitable slowness these people are going to experience from being unevenly associated with the APs?
    We can't re-write the application so we are looking for a wireless solution.
    Anyone hear about how other organizations have dealt with this type of situation?
    I'll be glad to supply more details if I am not being clear in my description of the problem.
    Thanks in advance. All responses will be rated.
    Paul

    This is the functionality which is missing in H-REAP: Client and Network Load Balancing
    "Radio Resource Management (RRM) load-balances new clients across grouped lightweight access points reporting to each controller. This function is particularly important when many clients converge in one spot (such as a conference room or auditorium) because RRM can automatically force some subscribers to associate with nearby access points, allowing higher throughput for all clients. The controller provides a centralized view of client loads on all access points. This information can be used to influence where new clients attach to the network or to direct existing clients to new access points to improve wireless LAN performance. The result is an even distribution of capacity across an entire wireless network.
    Note: Client load balancing works only for a single controller. It is not operate in a multi-controller environment."
    I suppose if we limit the number of users that can associate with a particular AP then we will achieve some client load-balancing. Though a hard limit on the number of end-users will also lead to situations where some end users will not be allowed any access.

  • Extend TCP Proxy vs. Client load balancing

    I am unclear how proxy and client load balancing interact with respect to custom address providers. If I define my own address provider, and I do NOT set the load-balancer parameter to client in the client configuration, will the proxy still do load balancing of connections as described in http://docs.oracle.com/cd/E24290_01/coh.371/e22839/gs_configextend.htm#BEBCICDA ?
    Edited by: user5179040 on Mar 23, 2012 9:43 AM

    Hi,
    The <load-balancer> element is only configured in the <proxy-scheme> and not at the client side. This parameter "proxy" dictates the proxy to use the specified strategy for load balancing client connections across proxies. The parameter "client" offloads the responsibility of load balancing to client across proxies or randomly select proxies.
    I am unclear how proxy and client load balancing interact with respect to custom address providers. If I define my own address provider, and I do NOT set the load-balancer parameter to client in the client configuration, will the proxy still do load balancing of connections as described in http://docs.oracle.com/cd/E24290_01/coh.371/e22839/gs_configextend.htm#BEBCICDA ?
    Hope this helps!
    Cheers,
    NJ

  • CSS on multiple subnets and separate load balancing

    Hello,
    I've a situation where I need to load balance incoming clients on subnet A to 3 real servers on subnet B - no problems there.
    But I also need to load balance different clients on subnet C to 3 other servers on subnet D and clients on subnet E to 2 servers on subnet F.
    Basically I want to use the CSS for 3 different load balancing operations.
    Rather than using 3 separate CSS11503s can I do all this with multiple VLANs on the LAN switches and 1 CSS?
    Any help appreciated
    Regards Tony

    you can have as many vlan as you want.
    So yes you can do what you want.
    Just be aware that the CSS can route as well between those vlans, so if you separation between them you may have to use ACL.
    Gilles.

  • Any concern on persistent search through a load balancer?

    We have access manager 7 installed which make use of persistent search. My understanding is that persistent search required to maintain a connection so that the server can refresh/update the client whenever entry in the result set changed. If we configure the system to connect to ldap through load balancer, will that cause any problem? What will happen if the load balancer refresh connection after a period of time? Or , if the original ldap server failed and the load balancer try load balance the client to another ldap server, will the persistent search still works?
    Also, if the ldap server that the persistent search initially established connection with crashed, will the client get error message and in that case, is it the client's responsibility to re-run/retry the persistent search with other failover ldap server?
    Thanks,

    Your best bet, even when using a hardware load balancer, is to front your DS instances with a pair of load-balanced Directory Proxy Servers. This way, you have physical redundancy at the load balancer level, and intelligent LDAP-aware load balancing at the proxy server level. DPS 6 is very nice in that you can split binds, searches, and updates amongst several backend DS instances, and the connection state is maintained by the proxy, not the DS instance (i.e. if an instance fails, you really shouldn't be forced to rebind, the proxy fails-over to another DS for searching).
    We have our Directory Servers on a pair of Solaris 10 systems, each with a zone for a replicated Master DS, and another zone each for a DPS instance. The DPS instances are configured to round-robin binds/searches/updates/etc. among the DS master zones. This works out very well for us.

  • Enabling server-side load balancing with Oracle 10g RAC

    Hi, all.
    I created a RAC database on vmware by using DBCA.
    By default, remote_listener and tnsname.ora have been configured.
    The followings are part of dba_services from my RAC database.
    NAME, GOAL, ENABLED, CBL_GOAL
    SYS$BACKGROUND NONE NO SHORT
    SYS$USERS NONE NO SHORT
    DEVDB NO LONG
    By default(created by DBCA), server-side load balancing is enabled for
    the RAC database, DEVDB??
    --> GOAL of DEVDB is null, but CBL_GOAL IS LONG.
    After creating a RAC db by using "DBCA", do I need to do something
    for server-side load balancing??
    If so what I need to do?? How can I test it??
    And what is beneficial in enabling server-side loadbalancing ??
    Is it good in the real world??
    Thanks for reading.
    Best Regards.
    Message was edited by:
    user507290

    [urgent] 10gR2 RAC Load Balancing Configuration (client and server side)!!
    [urgent] 10gR2 RAC Load Balancing Configuration (client and server side)!!
    Virag

  • WLC - Aggressive Load Balancing?

    Hello,
    The Wirless LAN Network bulit is as follows -
    1. 1 x 4404 WLC
    2. 40 x LWAPP 1131AG Access Points
    3. Windows Clients used by the Laptop Clients.
    4. Only one Wireless VLAN across the Capmus network - hence AP's, WLC & Clients are all in one VLAN / IP Subnet.
    5. No Access Point Group is created.
    6. Aggressive Load Balancing is enabled allowing 15 Clients as max connection per Access Point.
    Problem facing -
    1. Tried configuring the Aggressive Loadbalancing allowing only 2 x Clients per AP. But noticed that the 3rd Client connecting to the same AP as of the previous 2 Clients have connected. 3rd client is not associating to a different AP which is nearby.
    Please can one help me, if i'm configuring & testing Aggressive Load Balancing in the right way!
    Regards,
    Keshava Raju

    AMR is on target. In fact I just completed 20 hours worth of testing with variuos clients with ALB for a white paper I am doing. Code 17 isnt honored by most clients and is only sent 1 time from the AP. The clients will contiue to attempt to associate to the AP and the AP will allow them on.
    Here is a peek of my white paper "still in draft"
    WLC - Cisco WLC Aggressive Load Balancing; What is it and where did it go in 6.0!
    I've spent the majority of my WLC experience at code level 4.2. Not by choice really, more
    based on the fact that 4.2 is pretty darn stable and it is the only safe harbor to date for the Cisco WLC. Healthcare and Enterprise enviroments are typically slow to move on upgrades, especially when things are operating fine. 
    Since my latest project involves the deployment of hundreds of Cisco 1142s @ location grade, it required that I move to later code to support the 1142 access points. After much research, conversations with our
    local Cisco Wireless SE, conversation with peers at other healthcare organizations, and direct contact with the aware team I had decided that 6.0.188.0 was a release that was of great interest.
    As I start to get fimilar with the new code I am starting to see that things got moved around a little. One of the items is Aggressive Load Balancing. If you aren't fimilar with Aggressive Load Balancing (ALB) you definitly need to be and let me share why.
    First lets look at what ALB is and how it works and then we will dive into the differences between the 4.2 code and the new options 6.0 gives us. ALB when enabled, allows the Cisco WLC to load balance wireless clients on access points that are joined to the same controller. “Key word here – same controller”. You can configure the load balancing window globally in the controller. What is the load balancing window you ask? Well is the maximum number of clients that should be allowed on the access point BEFORE it will start to load balance.
    Lets assume for a moment you have an access point with 5 clients already attached. When client #6 sends association request to the access point the access point will kindly respond with an associaton response frame with the reason code of 17. The wireless client will see reason code 17 in the association response and will kindly find other access points to associate with. However, some devices will ignore this frame and yet still continue to try and associate to the access point. Note: The Cisco WLC will ONLY send 1 reassociation frame with a reason code of 17. It doesn’t flood the medium / client with multiple frames.
    Its up to the client to honor this information and move on. But I can tell you from my experience and testing this isn’t always the case.
    By default, 4.2 and 6.x both have a load balancing window of (5). Lets look at an example.
    The window setting controls when aggressive load−balancing starts. With a window setting of five, for
    example, all clients after the sixth client are load−balanced.
    I know, what is the reason code talk, right. Lets cover this as well. If you dive into the 802.11 frames you will see “Reason Codes”. When a client sees the reason code of “17”, it indicates to the client that the access point is busy and the client should look else where.
    yada yada yada
    I will post the complete paper on my site: my80211.com in the next week or so ...

  • CSS Citrix CAG Load Balancing

    Hi,
    I'm looking to get an opinion as to whether we should see even load balancing over two services.  The content rule is configured as follows :-
    content secure_cag
      add service citrix_cag_1
      port 443
      protocol tcp
      vip address 10.80.2.150
      balance srcip
      add service citrix_cag_2
      sticky-inact-timeout 240
      flow-timeout-multiplier 1800
      active
    Services :-
    service citrix_cag_x
      keepalive type tcp
      keepalive port 443
      ip address 10.200.16.18
      active
    At present we only have around 40 users using it but at times we are seeing a very uneven distribution of sessions, as much as 80% on one server.  Do we have too few users to see effective load balancing? Maybe our long timeout settings are breaking load balancing?
    Thanks for any insight anyone can share.

    Hi Chris,
    You might want to try balance leastconn for your balancing method.  Also, note that you are not currently configured for sticky, so the sticky timeout you have configured isn't doing anything.  Do you require sticky?  If you do not require sticky, then leastconn should give you the best distribution across services at any given point in time.  Adding sticky, such as with advanced-balance sticky-srcip, will skew load balancing as clients become stuck to one service.
    Hope this helps,
    Sean

  • WLC Load Balancing Threshold

    I am trying to understand how the load balancing threshold is calculated but I am finding conflicting information, even withing Cisco's own documentation. I would be grateful if anyone could help.
    Cisco's latest Wireless LAN Controller Configuration Guide for software release 7.0.116.0 (April 2011) contains the following information for configuring Wireless > Advanced > Load Balancing Page (emphasis mine):
    In the Client Window Size text box, enter a value between 1 and 20. The window size becomes part of the algorithm that determines whether an access point is too heavily loaded to accept more client associations:
    load-balancing window + client associations on AP with highest load = load-balancing threshold
    In the group of access points accessible to a client device, each access point has a different number of client associations. The access point with the lowest number of clients has the lightest load. The client window size plus the number of clients on the access point with the lightest load forms the threshold. Access points with more client associations than this threshold is considered busy, and clients can associate only to access points with client counts lower than the threshold.
    Option 1
    The formula shown is correct (load-balancing window + client associations on AP with highest load = load-balancing threshold). If so, this would mean that if you had a window size of 5 and the AP with the highest load at the time of calculation was 15, the threshold would be 18. However, as no APs have 18 associations then this threshold would never be reached. Even if an AP reach 18 associations, the next client trying to associate would trigger another calculation for the threshold which would be 21 (3 + 18) and so still, this threshold would never be hit.
    Option 2
    The description in the paragraph below is correct (The access point with the lowest number of clients has the lightest load. The client window size plus the number of clients on the access point with the lightest load forms the threshold). This sounds much more sensible to me. In this case, the window size was 3 and the AP with the lowest number of associations already had 7 clients associated, the load balancing threshold would be 10 i.e. no load balancing would occur until a client tried to associate with an AP which already had at least 10 clients associated.
    Option 3
    I have seen many descriptions on forums etc of the load balancing threshold being essentially the Client window size, i.e. if the client window size is 3 then load balancing will kick in when a client tries to associate to an AP with at least 3 clients already associated. This doesnt match the above documentation unless the AP with the least number of clients associated doesnt have any associated clients i.e. 0 clients.
    Questions
    I think Option 2 is the correct description of load balancing and the formula given stating use of the AP with the highest load is a typo (albeit still not corrected in the latest documentation). Am I correct?
    The problem with using the option 2 method of calculating the load threshold is that you will be unnecessarily performing load balancing in an environment where some of your APs do actually have zero clients associated, unless you set the window size to somehing close to 10.
    I read here http://www.perihel.at/wlan/wlan-wlc.html#aggressive-load-balancing that when calculating the load threshold, it only accounts for the 8 'best' APs for a given client. In other words, if you have 60 APs on your campus but only 20 are visible to the client, the controller will only perform its load threshold calculations bases on the 8 APs which have the best signal to the client. This would ,ake sense as there is no point setting a load threshold based on the lightest loaded AP which is not even within 'reach' of the client. Is this correct as I can not find any other documentation which supports this?
    Thanks in advance for your help with this.

    Interesting, the config guide contradicts itself in the same paragraph.....    I thought maybe we had two different documents with different explanations.  I don't see any open documentation bugs asking to correct this, but I swear I've heard discussion on this in the past.......
    First off:  Option #3 was the "old way". I think it changed in 6.0.    If you had a threshold of 5, then as soon as you had 5 clients on an AP it would reject the association (3 times and then let them on the 4th attempt).  Now its a sliding window/scale.
    Option #1 I think is completely wrong. As you described, how in the world would you ever surpass the threshold if the highest AP + the window is what you have to beat to load-balance....?    RIght, that just doesn't make any sense to me.....
    Option #2, the way you explain it is correct to my understanding...
    Your question #3 is also correct (not sure if it is Top 8 or based on an RSSI threshold though.)
    The idea is that you don't want some AP in a remote office with 0 clients being your starting point.   So I believe that it is based on the top X candidate for your client.    If your client has 4 viable candidates (lets just say -70 or better), and one of those APs has 5 clients and the rest have 15, I'd expect loadbalancing to try to get you to the 5 client AP if your window size was ~10......  something like that anyhow... 

  • Load-Balancing between Foreign and two Anchors

    Hi, we have two foreign controllers (one active, one standby) and two anchor controllers. All APs are connected to the active foreign controller. The layer 3 networks for the wlan clients on both anchors are different for the same SSID. SSID: Internet, anchor 1: Subnet A, anchor 2: Subnet B. So when a client is getting anchored to Anchor 1, the clients will get an ip from subnet A and when the client is getting anchored to anchor 2, the client will get an ip from subnet B.
    This is so far not a big problem because we only have a few accesspoints in some rooms. But what will happen, when we have a full covered wlan and the client roams from one AP to the other AP? Is there a possibility, that the client will anchored to a different anchor while roaming? I think this will result in a lack of connectivity because without a real disconnect the client will not ask for a new IP address.
    Other question: Is it possible to disable this load-balancing between anchor controllers? Or can i make a client sticky to only one anchor as long as an access-session is established?
    All controllers are 5760 with 3.3.3 software.

    Hi acontes, 
    It's an interesting question. 
    In this case, if all AP's are on WLC-A and there is no possibility that an L3 inter-subnet roam will occur between WLC-A and WLC-B, I would just forward WLC-A to Anchor A and WLC-B (in the event of fail over) to Anchor B (if Anchors reside on different subnets). If you must specify Anchor A and Anchor B on each WLC for redundancy purposes, it's important to understand the guidelines and limitations with regard to Foreign / Anchor Design.  
    As Scott mentioned, the limitation with Anchoring design is that there is no primary / secondary configuration for an Anchor on the Foreign WLC.
    If WLC-A has two entries (1) for Anchor-A and (2) for Anchor-B, the EoIP tunnels are establish and load-balancing occurs in a round robin fashion.
    Keep in mind the following with regard to guest N+1 redundancy:
    •A given foreign controller load balances wireless client connections across the list of anchor controllers configured for the guest WLAN. There is currently no method to designate one anchor as primary with one or more secondary anchors.
    •Wireless clients that are associated with an anchor WLC that becomes unreachable are re-associated with another anchor defined for the WLAN. When this happens, assuming web authentication is being used, the client is redirected to the web portal authentication page and required to re-submit their credentials.
    Since traffic is transported at Layer 2 via EoIP, the first point at which DHCP services can be implemented is either locally on the anchor controller or the controller can relay client DHCP requests to an external server. Since the IP address directly correlates to the DMZ subnet or the interface where the traffic egresses, it is possible for some clients to get IP's from both Subnet A or Subnet B in the event that WLC-A is building EoIP to both anchors.
    1) What happens if my clients roam?
    Nothing... since all AP's are on WLC-A, it's Intra-Controller Roaming
    Each controller supports same-controller client roaming across access points managed by the same controller. This roaming is transparent to the client as the session is sustained, and the client continues using the same DHCP-assigned or client-assigned IP address. The controller provides DHCP functionality with a relay function. Same-controller roaming is supported in single-controller deployments and in multiple-controller deployments.
    Would it be better to choose the same DHCP Pool on both anchors?
    It's probably better to have redundant anchors on the same subnet, but it's not required. 
    3) How would you design this :-)
    WLC-A <--EoIP--> Anchor A (DHCP Pool A)
    WLC-A <--EoIP--> Anchor B (DHCP Pool A)
    It's important to remeber what Scott mentioned about the lack of a primary / secondary relationship. If multiple controllers are added as mobility anchors for a particular WLAN on a foreign controller, the foreign controller internally sorts the controller by their IP address. The controller with the lowest IP address is the first anchor. For example, a typical ordered list would be 172.16.7.25, and 172.16.7.28. If the first client associates to the foreign controller's anchored WLAN, the client database entry is sent to the first anchor controller in the list, the second client is sent to the second controller in the list, and so on, until the end of the anchor list is reached. The process is repeated starting with the first anchor controller.
    If any of the anchor controller is detected to be down, all the clients anchored to the controller are deauthenticated, and the clients then go through the authentication/anchoring process again in a round-robin manner with the remaining controller in the anchor list. This functionality is also extended to regular mobility clients through mobility failover. This feature enables mobility group members to detect failed members and reroute clients.

  • ASA and vpn load balancing

    Hi,
    I am configuring 2 ASA5540 for internet trafic inside to outside ,
    outside to inside (web,smtp) but also vpn load balancing for client to site , site to site and webvpn.
    In the doc I can configure them for internet trafic as Active/Standby or Active/active.
    for vpn : I can use vpn load balancing
    But no information if I want to use the active/passif and vpn load balancing together.
    Any thoughts on which way to go? what is the best thing to do ?
    Regards

    Hi,
    I think that you cannot use an Active/Active configuration for VPN connections as it is stated on Cisco's documentation: "Note: VPN failover is not supported on units that run in multiple context mode as VPN is not supported in multiple context. VPN failover is available only for Active/Standby Failover configurations in single context configurations" available at http://www.cisco.com/en/US/products/ps6120/products_configuration_example09186a0080834058.shtml
    Hope it helps

  • Load balancing in 1250 Series (Autonomous) AP

    Is it possible I can load-balance my client ? 
    In our current setup when one AP down rest of the client move to other AP, we need to move them manually or connect/disconnect. I can apply max-association on SSID but is there way I can do it automatically ? 

    Not remotely possible. 
    The "final" decision to join which AP and/or which radio falls on the wireless client.  If the wireless client wants to associate to the furthest AP then there's nothing you can do. 
    One way of minimizing this from happening is to ensure your wireless clients are running the latest drivers.   On the AP-side, you could also try to disable the low-speed data rates (from 11 Mbps and below, make 18 Mbps as Mandatory and the rest are Supported).  

  • Remove load balancing...

    Oracle 10gR2 RHEL 4 AS 64bit
    I wanted to know is there a way to remove load balancing in RAC? Would I just set the LOAD_BALANCE parameter to OFF in the TNSnames.ora file? Would that take care of it? The problem is occurring because developers are trying to upload images into the database (using jumploader) and the file is stored on the first node. What happens is during the upload a procedure searches for the file on the second node (when it is actually on the first node) and then an error is returned. Below are my current TNS entries:
    PROD =
    (DESCRIPTION =
    (LOAD_BALANCE = ON)
    (FAILOVER = ON)
    (ADDRESS = (PROTOCOL = TCP)(HOST = node1-vip)(PORT = 1521))
    (ADDRESS = (PROTOCOL = TCP)(HOST = node2-vip)(PORT = 1521))
    (CONNECT_DATA =
    (SERVER = DEDICATED)
    (SERVICE_NAME = TAF)
    (FAILOVER_MODE =
    (TYPE = SELECT)
    (METHOD = BASIC)
    (RETRIES = 180)
    (DELAY = 1)
    TAF =
    (DESCRIPTION =
    (LOAD_BALANCE = ON)
    (FAILOVER = ON)
    (ADDRESS = (PROTOCOL = TCP)(HOST = node1-vip)(PORT = 1521))
    (ADDRESS = (PROTOCOL = TCP)(HOST = node2-vip)(PORT = 1521))
    (CONNECT_DATA =
    (SERVICE_NAME = TAF)
    PROD2 =
    (DESCRIPTION =
    (ADDRESS = (PROTOCOL = TCP)(HOST = node2-vip)(PORT = 1521))
    (CONNECT_DATA =
    (SERVER = DEDICATED)
    (SERVICE_NAME = PROD)
    (INSTANCE_NAME = PROD2)
    PROD1 =
    (DESCRIPTION =
    (ADDRESS = (PROTOCOL = TCP)(HOST = node1-vip)(PORT = 1521))
    (CONNECT_DATA =
    (SERVER = DEDICATED)
    (SERVICE_NAME = PROD)
    (INSTANCE_NAME = PROD1)
    )

    In this case, changing the tnsnames.ora file alone won't do it. RAC has two independent load balancing sides, client side (tnsnames level) and server side (listener level). The client side is simply a random (yep, random) picking of a listener on an instance to go to. On the server side, however, a given listener can then hand that connection off to another listener on a less loaded machine. Even if you connect via tnsnames to a specific instance, you may be connected to another machine.
    Your best bet would be to use services, you can create a service in the cluster that can connect to a specific node with a different service name than "standard" connection. I have used this for reporting instances on a specific node cluster. You then create a new tnsnames alias to point to that new server. Note. If you use the SERVICE_NAMES parameter within your database this will not work, as anything pointed to by it will override the service names.
    An alternate idea would be to nfs mount your filesystem that contains the pictures to each of your remaining rac nodes.
    Cheers
    Jay Caviness
    http://www.grumpy-dba.com

Maybe you are looking for