HP Switches Controlling Network Traffic Hewlett-Packard switches offer an array of features designed to provide increased network performance with a minimum of complication and administration. Among features that fit this description are Switch Meshing, Automatic Broadcast Control, support for IGMP and the possibility of replacing routers in a local LAN environment with simpler HP switches. This application note examines ways to use Hewlett-Packard switches to limit and control network traffic, especially the broadcasts that often are among the biggest challenges facing network administrators. Topics covered are: Implementing Automatic Broadcast Control in both IP and IPX environments IGMP details Considerations needed when replacing a pre-existing router with a switch in the local LAN environment Spanning Tree Protocol in a NetWare environment Switch Meshing is covered in a separate white paper available at http://www.hp.com/rnd/techlib/techlib.htm. Broadcast traffic is a normal and necessary function in all network protocols, including IP and IPX. Broadcast traffic must, however, be controlled as the network gets larger. Automatic Broadcast Control (ABC) is a feature that allows HP switches that implement it to provide broadcast isolation between the different segments of the switch without requiring routers. Because the functioning of ABC is different depending on whether the environment is IP or IPX, both circumstances are described. Broadcasts Raise Performance Issues One of the biggest issues in contemporary networking is the control of broadcasts. Many network devices and processes rely on broadcasts to discover and monitor other devices and processes on the network. Much of the broadcast traffic on a typical network is generated by workstations, servers and other devices trying to locate each other.
Layer 2 Switch ARP Broadcast Hub Hub Hub Hub Printer WS WS Printer WS WS Printer WS WS Printer WS WS Figure 1. Layer 2 Switches Forward Broadcasts to All Ports Typical Layer 2 switches flood broadcasts, meaning they forward the broadcast packets out every port except the one on which it arrived, as when the workstation connected to the far right hub sends out an ARP request as shown in Figure 1. When a large network is interconnected with Layer 2 switches, large amounts of broadcast traffic can result. Broadcast traffic has a twoedged effect on a Layer 2 network: 1. Broadcasts increase network utilization of every network segment. Unlike unicast traffic, which is segmented by the switch, broadcast traffic is flooded to every network segment or collision domain and therefore uses network bandwidth on all of the segments. 2. Broadcasts adversely affect the performance of every workstation and server on a Layer 2 network. Network interface cards (NICs) are unable to evaluate the applicability of broadcast packets and must pass them to the upper layers of the networking software for evaluation. Even though it is likely that the packet is not applicable to this workstation, this evaluation forces a CPU interrupt and steals CPU cycles. The demands of broadcast traffic alone can affect the performance of other applications, especially on systems with limited CPU and memory resources. Broadcast storms are another serious problem affecting Layer 2 networks. A broadcast storm is a condition in which broadcast packets propagate and multiply on every segment of the network, using all or most of the available bandwidth and effectively shutting down the network. Broadcast storms are generally caused by a topology loop, an addressing problem, or a faulty network device. Typical Layer 2 switches provide no protection from broadcast storms because of their inability to control broadcast traffic. Routers: The Traditional Tool for Broadcast Control The traditional method for controlling broadcasts is to place routers at the center of the network, where they segment network traffic and separate the network into multiple broadcast domains. Routers inherently prevent broadcasts from crossing from one network segment to another, thus limiting the effects of the broadcasts to a single broadcast domain. In this role at the center of the network, routers raise the network complexity, requiring more sophisticated configuration, detailed administration, and often, higher equipment costs. Page 2 of 15
Controlling Traffic with Switches It is not practical to replace a router with a basic Layer 2 switch for two reasons: broadcast control is lost, and security is lost. If a router is replaced with a basic Layer 2 switch, the network will be flattened; that is, the network will become a single broadcast domain where all devices will be able to hear broadcasts from all other devices. In addition, the basic Layer 2 switch cannot provide the highly sophisticated security capabilities of the router. However, on networks that do not require sophisticated levels of security, HP switches with Layer 3 traffic controls can be used to replace routers. In these cases, the switch uses the Layer 3 traffic controls to provide broadcast control similar to that of a router. In general, the Layer 3 traffic control features use information from the Layer 3 header to make intelligent forwarding decisions. The Layer 3 traffic controls do not enable HP s ProCurve switches to replace routers in every instance. Instead, it is advisable to consider these switches as possible alternatives to routers in situations where the high-level of security offered by routers is not necessary. In other words, the HP ProCurve switch family can often provide sufficient control over network traffic without the overhead and cost associated with routers. Routers, Routing Switches Control HP Switches with Layer 3 Traffic Controls Layer 2 Switches Complexity Figure 2. Control and Complexity Increase Together As Figure 2 suggests, HP switches with Layer 3 traffic controls offer a significantly higher level of control with little additional complexity. By contrast, routers or routing switches offer almost complete control over network traffic, but are notoriously difficult to configure and maintain. On the other extreme, switches that feature only Layer 2 technologies provide almost no control over traffic issues such as broadcast. The current generation of HP switches provides this balance between complexity and control by managing broadcasts and multicasts. Multicast transmissions are becoming increasingly common as video-conferencing and other multimedia applications pervade the business world. The technologies that form Layer 3 traffic controls are: Automatic Broadcast Control (ABC), which suppresses broadcasts by participating in the IP and IPX broadcast protocols for host and route discovery. Protocol Filtering, which can restrict traffic between segments on the basis of the protocols used. HP ProCurve switches enable the configuration of filters based upon the AppleTalk, IP, and IPX protocols. Filtering can also be done on these Layer 2 protocols: SNA, DEC LAT and NetBios. Page 3 of 15
ƒ Internet Group Multicast Protocol (IGMP), an Internet standard that enables routers and Layer 3 switches to direct multicast traffic to the appropriate nodes without affecting the performance of other nodes. Taken together, these features provide virtually the same level of broadcast control as routers. Automatic Broadcast Control Automatic Broadcast Control (ABC) helps to conserve bandwidth and processing power for IP traffic, IPX traffic, or both within a broadcast domain. The HP switch greatly reduces broadcasts by either stopping them at the switch, or by having the broadcast requests answered directly by the switch rather than sending the broadcast on. ABC controls broadcasts through two mechanisms: broadcast rate throttling and IP/IPX broadcast reduction. ƒ Broadcast Rate Throttling causes the switch to drop any broadcasts that exceed a configured percentage threshold, which guards against broadcast storms. This function, with the threshold set on a port-by-port basis, is independent of IP or IPX and not discussed any further in this application note. ƒ IP/IPX Broadcast Reduction significantly reduces the propagation of broadcasts generated by IP and IPX clients and servers. Examples of such broadcasts are IP ARPs and RIPs, and IPX NSQs, RIPs, and SAPs. This concept is illustrated in Figure 3. In the network on the left, the workstation on the left sends out an ARP request looking for the server on the left. This ARP request is a broadcast, so it is transmitted throughout the network. The server in the upper left detects that it is the node being looked for and replies back to the originating workstation. In the network on the right, the switch with ABC gets the ARP request from the workstation, looks up the response it is own ARP cache and sends the reply directly back to the workstation. The broadcast stops right at the switch. Server Server Server Server Server Server Switch HP Switch with ABC Switch Switch HP Switch with ABC HP Switch with ABC WS WS WS WS WS WS Figure 3. Activating Automatic Broadcast Control How ABC Works When ABC is enabled on an HP switch, the switch snoops or listens to solicited or unsolicited IP and IPX address resolution responses and caches them. Subsequently, when the switch receives a broadcast address request, it sends the cached response and drops the broadcast request instead of flooding it. ABC performs the following functions: Page 4 of 15
1. With ABC for IP enabled, the switch builds an ARP cache, and gives a proxy ARP reply for end nodes whose IP-to-MAC address mapping the switch has learned through snooping on previous ARP requests. 2. With IP RIP control enabled, the switch forwards IP RIP broadcasts only to other ports on which the switch has received IP RIP broadcasts. 3. With ABC enabled for IPX, the switch builds IPX route and service tables, and gives a proxy reply to "GetNearestServer" and "GetLocalTarget" IPX broadcast requests. 4. With IPX RIP/SAP control enabled, the switch forwards IPX RIP and SAP broadcasts only to other ports on which it has received IPX RIP and SAP broadcasts. Automatic Broadcast Control is available in the HP ProCurve Switch 1600M, the HP ProCurve Switch 8000M, the Hewlett-Packard AdvanceStack Switch 800T, and the AdvanceStack Switch 2000. Although we are discussing ABC in the context of replacing routers, ABC can also provide significant broadcast reduction benefits when implemented in a router-centric network as shown in Figure 4. Implementing ABC in this case requires only that ABC is enabled on the switch. No configuration changes are required on any other devices. We will discuss the configuration changes required to replace a router with a switch later in this section. Figure 4. ABC in a Router-centric Network Protocol Filtering It is not unusual for different workgroups of a network to have different networking needs. Sometimes, workgroups require networking protocols that are not used by other workgroups. For example, it is not uncommon for graphic artists and graphics departments to use Apple Macintosh computers, which typically use the AppleTalk networking protocol. Similarly, it is not uncommon for some workgroups to require Novell NetWare services, while others rely solely upon Microsoft Windows NT or upon the various versions of UNIX and the TCP/IP protocol suite. Page 5 of 15
In flat Layer 2 networks, it can be impossible to prevent the broadcast traffic from one workgroup from impacting other workgroups. This problem can be especially acute in the case of AppleTalk and Novell, which rely heavily upon broadcasts to establish connections. Figure 5. Protocol Filtering Protocol filtering, which is configured on a per-port basis, can be used to stop the unwanted broadcasts of a particular Layer 3 protocol from being flooded to other parts of the network, as depicted in Figure 5. HP ProCurve Switches Filter Protocols to Limit Traffic The HP ProCurve Switch 8000M and HP ProCurve Switch 1600M, as well as the HP AdvanceStack Switch 2000 and 800T, address the issue of protocol proliferation by providing Layer 3 filtering mechanisms for traffic in IP, IPX and AppleTalk protocols. The Layer 2 protocols DEC LAT, SNA, NetBios can also filtered. The switches can be configured to forward or drop frames in any of these protocols on a per-port basis. In addition, filters for specific multicast addresses can be defined to forward or drop multicast traffic on a per-port basis. These filtering options enable administrators to streamline traffic. If, for instance, a graphics workgroup is using AppleTalk, but the traffic is not desired elsewhere on the network, administrators can configure an AppleTalk filter so that no AppleTalk traffic will be forwarded to other workgroups. Because of the per-port options, administrators can also assure that the AppleTalk traffic will be forwarded to some workgroups and not to others, enabling a high degree of customization in network resource allocation. Administrators can use the same technique to configure filters for IP and IPX traffic. In addition to protocol filtering, the HP ProCurve Switches provide port filters. This feature provides a convenient method for ensuring the security of some network resources. For example, if a file server is connected to a specific port, file server traffic can be excluded from other ports, preventing users on those ports from gaining access to the server. Page 6 of 15
IP Multicasting and IGMP Unlike the majority of today's point-to-point network applications, emerging multimedia applications such as LAN TV, collaborative computing, and desktop conferencing depend on the ability to send the same information from one host to many hosts or from many hosts to many hosts. The rising need for this type of multipoint communication presents an interesting challenge for networks. For example, email systems can use multipoint communication to deliver directory updates to multiple servers simultaneously. Mail servers can update their databases in one session instead of having to establish multiple point-to-point sessions between systems. Some multimedia applications can also use multipoint services to facilitate the delivery of CD-ROM, live TV video feeds to multiple systems across the network, or multipoint video conferencing. Figure 6. IGMP (IP Multicast) To help networks and network administrators cope with the rising tide of multicast traffic, the IP community developed IP multicasting and the Internet Group Multicast Protocol (IGMP). Multicasting sends information packets to a pre-defined subset of network devices. For this reason, it has significant advantages over both broadcasting and unicasting. Multicasts are far more efficient than broadcasts because they are sent to a subset of network devices instead of all devices. Unicasts, of course, send packets to single nodes. For reaching multiple users, they are much less efficient than multicasts because a separate copy of each packet is sent to each destination. Some outstanding benefits of IP multicasting are: 1. The sender only has to send out one copy of the information packet instead of many. 2. The information is delivered in a more timely, synchronized fashion because all destinations receive from the same source packet. 3. Multicasting can be used to send information to destinations whose individual addresses are unknown (similar to a broadcast). 4. Multicasting reduces the overall number of packets on the network because one multicast packet is sent instead of many unicast packets. Page 7 of 15
Managing Multicasts with IGMP Layer 2 switches treat multicast packets exactly as they do broadcast packets. This means that in a network connected by Layer 2 switches, multicasts are flooded to every device on the network. In an environment rich in multicast types of applications, this could affect the performance of all hosts on the network. IGMP was developed to address this problem. IGMP, an internal protocol of the Internet Protocol (IP) suite, provides a means to automatically control and limit the flow of multicast traffic through the network. Applications and networking devices that implement IGMP work together to eliminate multicast traffic on segments that are not destined to receive this traffic. Implementing IGMP with HP ProCurve Switches All of the currently available managed HP ProCurve and AdvanceStack switches support IGMP. All of these switches are IGMP queriers, independently discovering multicast groups when multicast routers are not present. They also can learn which of their interfaces are linked to other hosts and multicast routers. Configuration is simple just click a single checkbox and IGMP is enabled. Most of the HP switches also can be configured through a single checkbox to promote multicast streams that are part of an IGMP group to the high-priority queue. These packets move through the switch with the least amount of latency delay or wait time as possible. Layer 3 Traffic Control Summary Taken collectively, the Layer 3 traffic controls Automatic Broadcast Control, Protocol Filtering, and IGMP enable HP switches to assume the role traditionally fulfilled by LAN routers. Using switches in this way presents three clear advantages: 1. Higher speed. Switches generally are faster than routers and can prevent or alleviate the bottlenecks often caused by routers. 2. Lower cost. Because switches are simpler than routers, they are less expensive to buy, configure, and maintain. 3. More efficient use of existing routers. Routers that are currently being used to segment LANs can be re-deployed more effectively at the edges of networks, where they can connect LANs to WANs and to the Internet. For those environments that require a high level of security or precise control over traffic at Layer 3, however, a routing switch may be a better solution. Page 8 of 15
Replacing a Router with a Switch - IP Environments In general, when replacing a router with a switch at the center of a network, the most important of the Layer 3 traffic controls is ABC. Without ABC, the switch could not control broadcasts and would not be suitable for its new role. Consequently, before replacing a router with a switch featuring Layer 3 traffic controls, we must carefully examine the configuration options for the switch and for the network devices it will connect. IP Configuration Options Figure 7. Typical Router-centric Network In Figure 7, the objective of replacing the router with a switch in an IP environment is to increase the capacity of the network while keeping costs and complexity as low as possible. If we were to remove the router, replace it with a switch and make no other configuration changes, the end-nodes on different subnets would no longer be able to communicate because the router functioning as the default gateway would be gone. We must, then, make configuration changes so that the end-nodes on different subnets can communicate. There are three possible options for making the changes. The three options are directly related to the three pieces of information critical to a device s IP configuration: Option 1 IP Address: We can change the IP address of all the devices so they are all on the same subnet. Option 2 Subnet Mask: We can change the subnet mask of all devices to transform what were formerly separate subnets into a single subnet. This is called supernetting. Option 3 Default Gateway: We can change the default gateway of each end-node to point at the device itself. This is the least obvious approach and requires the most explanation. Option 3 is usually the best option of the three. Page 9 of 15
Option 1: Re-addressing the devices Figure 8: Alternative 1: Re-addressing This option is straightforward and easy to understand. All of the devices on the network (endnodes and connectivity devices) are assigned new addresses in a single address space large enough to contain them. After this change, all devices in this network will be able to communicate with each other through the switch. In general, this option is not acceptable because the impact of changing the addresses of device has far-reaching and sometimes unexpected results. In addition, it is difficult to get a single IP address space large enough to contain all current nodes and allow for future growth for all but the smallest networks. Option 2: Supernetting Figure 9: Alternative 2: Supernetting This option works only if all of the subnets on the router are contiguous. Figure 7 shows a router-centric network with contiguous addresses. If the subnet addresses are contiguous, it is possible to change the subnet mask and combine separate subnets into a single subnet. As mentioned previously, this is called supernetting. In this option, the subnet masks of all of the Page 10 of 15
devices (end-nodes and network connectivity devices) are changed. Figure 9 shows the network after changing the subnet mask. This option should be used with extreme care. All current subnet address ranges must be contiguous; any new subnets used in the future must also be contiguous to continue using the supernetting approach. Knowing what value to give the subnet mask can be confusing and should be done only by those who are very familiar with the subnet mask concept. Option 3: Changing the Default Gateway Figure 10: Alternative 3: Changing the Default Gateway When a device s default gateway is changed to point at itself, that is the default gateway IP address is the same as the device s IP address, the device ARPs directly for all devices on the network instead of only those on the same subnet. Figure 10 shows the network after changing the default gateway. Because they are broadcasts, these ARP requests can reach any other device on a network interconnected by switches. This is true even in cases where the Layer 3 addresses indicate that nodes are on different subnets. Consequently, the nodes can communicate directly without a router between them after changing the default gateway. Changing the default gateway is generally the most attractive of the three options. On a statically addressed network, the default gateway of each device must be changed. The cost of this effort must be weighed against the benefits of replacing the router with a switch. The cost of this effort can be reduced substantially in dynamically addressed networks by using Dynamic Host Configuration Protocol (DHCP). DHCP simplifies network addressing by automatically assigning addresses, subnet masks, and default gateways to network nodes when they are booted up. Changing the default gateway address on the DHCP server makes implementing this option simple. However, a Windows NT 4.0 DHCP server normally will not allow a device s IP address and default gateway to be identical without making changes to the registry. For details see: http://support.microsoft.com/support/kb/articles/q167/6/86.asp. A related DHCP discussion regarding superscopes (configuring for multiple subnets on the same DHCP server port) for Windows NT 4.0 can be found at: http://support.microsoft.com/support/kb/articles/q161/5/71.asp. Page 11 of 15
Automatic Gateway Configuration To make implementing this option simpler, HP has implemented a feature called Automatic Gateway Configuration (AGC) on its switches that support ABC. AGC enables switches to listen for DHCP responses. When the switch hears a DHCP response from a server, the switch modifies the assigned default gateway to be the same as the assigned IP address before relaying the response to the device. 2. Windows NT Server replies: IP Address: 15.15.15.98 Default Gateway: 15.15.15.1 Switch Server 3. Switch intercepts the reply and copies the IP address over Default Gateway assigned by the server IP Address: 15.15.15.98 Default Gateway: 15.15.15.98 1. Workstation sends DHCP request WS Figure 11: Automatic Gateway Configuration Any end nodes that are on the same switch segment as the DHCP server will not benefit from Automatic Gateway Configuration since those packets do not travel through the switch. It is recommended when using the Automatic Gateway Configuration feature that DHCP servers be located on switch segments without any end nodes. If the DHCP server itself allows setting the default gateway to the end node address it is recommended that the DHCP server be used rather than the Automatic Gateway Configuration in the switch. ABC Enables HP Switches to Resolve ARP Broadcasts When one of the options above has been implemented each node now ARPs for any node unknown to it. Unless the switch we ve put in the center of the network can resolve those broadcast requests, they will be flooded across the network. This is where ABC is helpful by enabling the HP switch to cache the ARP requests and resolve them directly, as shown in the network on the right in Figure 3. Now the devices can talk to each other through the switch and broadcasts are kept at a minimum. Setting Proxy ARP in any Remaining Routers In Figure 10, the devices connected by the switch will now ARP directly for devices on the WAN. This could be a problem because the router will block the broadcast requests from reaching the WAN devices. This will prevent nodes from finding any nodes on the other side of the router. Routers, however, have a feature called Proxy ARP that will resolve this problem. When the Page 12 of 15
devices on the switched network ARP for devices on the WAN, the router (with Proxy ARP enabled) responds to that request, and forwards the packets exactly as it would in the role of a default gateway. Proxy ARP must be enabled manually on most routers. This discussion also applies to any routers or routing switches left in the local LAN environment. Proxy ARP should be enabled on these routers as well. Replacing a Router with a Switch - IPX Environments The goal in replacing the router with a switch in the IPX environment is similar to IP. Devices should communicate directly through the switch and use the router only to access devices across the WAN. Figure 12 depicts a typical router-centric IPX network. WAN Network: 01F0A23 Router Server Network: 01F0A2F Network: 01F0A2B Server Switch Hub A B C D Figure 12: Typical IPX Router-centric Network How ABC Works in the IPX Environment IPX addressing is dynamic. A separate IPX network address is configured on each interface of the router. As opposed to IP, only the servers and routers must be configured with this address. Workstation network addressing is completely dynamic. Another configuration item in the IPX environment is frame (encapsulation) type. There are four Layer 2 frame types supported by IPX. Each network number in IPX can be locked to only one frame type. This feature can allow for four independent logical networks on one physical link, i.e. four network numbers, each using a different frame type. When multiple frame types are used, devices that use different frame types cannot communicate directly with each other; communication is only possible through a router or server. When an IPX node first comes up it broadcasts a Get Nearest Server (GNS) request (also known as a Nearest Server Query or NSQ) using the IPX network number 0 (zero). A router or IPX server responds back from its RIP/SAP table with the actual local IPX network number, network number of the nearest server that can service the needs of this node and the MAC address of itself. The HP Switch with ABC enabled builds its own RIP/SAP table and replies back to the node with the above information, except that the MAC address given is the MAC address of the actual nearest server, not the switch. At this point two things have happened: Page 13 of 15
The GNS, a broadcast packet, has gone no farther than the switch. The IPX node now has the MAC address of the nearest server and can communicate directly with this server through the switched network. ABC in the HP switches will handle each frame type separately and correctly, but being a Layer 2 switch, will not provide communication between devices using different frame types. That communication must still come through an IPX router or server. When replacing a pre-existing IPX router with an HP switch as in Figure 12, the network number and frame type configured on the local router interface must agree with the network number and frame type configured on any Netware servers on that interface. These are the only changes required. This new configuration is depicted in Figure 13. Figure 13: IPX Re-addressing Common Problems Seen in IPX Environments There are three common problem situations that come up in an IPX environment. While these problems are actually the result of illegal IPX configurations in the first two situations and Spanning Tree Protocol in the third, they usually show up when the switch is inserted into the network. This may lead the user to incorrectly think they were caused by the switch since that was the last thing that was done before the problem showed up. Replacing a Router With an HP Switch As mentioned above, replacing a pre-existing IPX router with an HP switch requires that all servers that are connected via the switch have the same network number as the interface on the router. Reconfiguring the servers for the correct external network number is all it takes. Each server retains its existing internal network number. Because IPX end nodes learn their network numbers dynamically, it may take a reboot of the end nodes before they learn the new network number. If the network numbers are not changed to agree, all servers and routers in the common domain will report errors on their consoles stating multiple network numbers are present. In addition, if ABC is enabled for IPX and the switch sees packets with different network numbers using the same frame type, the switch updates the network number it has for a particular frame type based on the last packet seen. This can occur up to five times in five minutes. At the sixth change, the switch locks on to the network number seen in the fifth change Page 14 of 15
and disables ABC for IPX, flooding all new GNS requests through it. (The switch locks on to a network number even though it has disabled ABC under IPX so that the switch can still respond to an IPX diagnostic packet, since this is a common way for IPX devices to discover other IPX devices on a network.) This sequence will be indicated in the event log of the switch. Once the servers have been configured correctly, ABC for IPX can be re-enabled on the switch. Same Network Number on Different Frame Types Another common problem is having the same network number show up on different frame types, which is an illegal state under IPX. This can be caused through incorrect configuration, or more commonly, when upgrading a server from Netware 3 to Netware 4 since the default frame type is different between these two versions. If this situation occurs, console errors will be seen on both IPX servers and routers in the environment. The situation is fixed by configuring all servers using the same network number to agree on a frame type. Unlike network numbers, which are automatic for end nodes, end nodes may also have to be re-configured for frame type if their frame type has been set via configuration. If an HP switch with ABC enabled detects the same network number on multiple frame types, it will lock on to the last frame type seen for a particular network number. If the frame type for a specific network number changes five times over a short period of time, the switch will disable ABC for IPX. (The switch locks on to a frame type even though it has disabled ABC under IPX so that the switch can still respond to an IPX diagnostic packet, since this is a common way for IPX devices to discover other IPX devices on a network.) Spanning Tree in the IPX Environment When replacing an IPX router with a switch the user can get the following message when powering up a PC, "A file server could not be found". This is a well-known issue given the following situation: 1. The PC is directly-connected to the switch 2. The PC is running Novell's VLMs or Client32 3. The switch has Spanning Tree Protocol (STP) enabled In this situation, when the directly-connected PC is powered on, the switch senses linkbeat on that port. This causes the switch to go through the four Spanning Tree states: blocking, listening, learning, and forwarding. It takes 30 seconds for the switch to complete that sequence and begin forwarding packets to and from that port. During those 30 seconds, Novell sends 3 requests for a server, then stops looking. By the time Spanning Tree completes its job, Novell reports that "a file server could not be found". Current versions of firmware (found at http://www.hp.com/cposupport/indexes1/ashubs.html) for the HP ProCurve Switches that support STP, include an enhancement to resolve this timing problem between Novell and STP. The enhancement allows users to configure Spanning Tree, on a port-by-port basis, in fast mode so that it does not go through the 4 states. Instead, for those ports configured in fast mode, Spanning Tree will immediately begin forwarding packets to and from the port. This allows Novell clients to communicate with the server as soon as the network card (NIC) is enabled. After that, the switch continues to listen for, and send, Spanning Tree packets on those configured ports. This protects the user who might inadvertently connect a hub or switch to that port and create a network loop - Spanning Tree will detect the loop after a short time, since the port listens for and sends STP packets on that port. Technical information subject to change without notice. Copyright 1999, Hewlett-Packard Company Page 15 of 15