Posted: February 18th, 2010 | Author: charlie | Filed under: Networking | Tags: ccie, ccna, cisco, networks | 2 Comments »
Cisco Flex Links gives network operators a simple, reliable, and more scalable method of layer 2 redundancy. The Spanning Tree Protocol (STP) is not destined for the scrap bin, but it will certainly fall out of favor with many enterprise networks.
Flex Links are a pair of layer 2 interfaces configured to act as a backup of each other. Configuring Flex Links is very simple, but it’s a manual process. Spanning tree can configure itself if you just enable it, albeit likely a sub-optimal configuration, but a working one nonetheless. Flex Links, on the other hand, require manual setup and layout of your layer 2 network. If you don’t want to leave anything to chance, then Flex Links are preferred over STP.
The benefits of FlexLinks include:
- simplicity, which equals stability.
- instant failover.
- rudimentary load balancing capabilities, so one link isn’t wastefully idle.
- load balancing works across switches in a stack, including port channels.
Flex Links’ primary operating mode is just like spanning tree: one on, one off. With per-VLAN spanning tree, a trunk port can have some VLANs enabled and some blocked at the same time, so on the surface it seems that STP is superior. In reality, you can configure Flex Links to load balance VLANs, and we’ll show you how shortly.
Configuration
Conceptually, you configure Flex Links by telling one link it’s the active link, and another that it’s the backup of that
Flex Links Design Map
primary (active) one. Without configuring VLAN load balancing, it will completely disable the backup, and if the active link goes down the backup will take over.
For example, to configure port gi1/0/1 as a active link, and gi1/0/2 as the backup, you’d run:
Switch# configure terminal
Switch(conf)# interface gigabitethernet1/0/1
Switch(conf-if)# switchport backup interface gigabitethernet1/0/2
That’s all there is to configuring the basic mode, which gets you failover but no load balancing. Before talking about load balancing, let’s take a look at preemption and “mac address-table move update.”
Preemption
Preemption, that is, the preferred port for forwarding traffic, is also configurable. This is most often used in combination with multiple links that have differing bandwidth capacities. If you wish to ensure that port 1, a primary port that has more bandwidth, will return to the active link when it comes back up, you would set: interface preemption mode bandwidth andswitchport backup interface preemption delay. The delay is used to set the amount of time (in seconds) to wait before allowing port 1 to preempt port 2 and begin taking over traffic again.
MAC Address-Table Move Update
Enabling the MAC address-table move update feature allows for rapid convergence when a primary link goes down and the backup takes over traffic forwarding duties. Without this feature enabled, neighboring switches may continue to forward traffic for a short time to a dead port, since they have learned MAC addresses associated with that link.
When move update is enabled, the switch containing Flex Links will broadcast an update packet to let other switches know what happened, and they will in turn un-learn that false MAC address mapping.
On the switch with Flex Links, simply configure:
Switch(conf)# mac address-table move update transmit
All switches, including ones with Flex Links, need to receive these updates. This is not enabled by default, so you’ll need to run the following command on all of your devices:
Switch(conf)# mac address-table move update receive
To see the status and verify that “move update” is enabled, run: show mac address-table move update. Checking the status of your Flex Links is much the same: show interfaces [interface-id] switchport backup.
Load Balancing
Flex Links should be configured such that both ports are forwarding traffic at the same time. This way, you get load balancing in addition to redundancy. The limitation is that only one port can be forwarding a single VLAN at a time. If we have VLANs 1-200, we need to choose which VLANs are forwarded primarily through which port. The most simple configuration, ignoring traffic requirements, would be that VLANs 1-100 use port 1, and VLANs 101-200 use port 2.
Before we get into configuring preferred VLANs, let’s talk about multicast. Multicast, of course, becomes an issue with this type of setup. If a port passed an IGMP join, and the switch is part of a multicast group, when the port goes down the switch will no longer be able to receive multicast traffic for that group. The quick fix is to make both Flex Links always be part of learned groups, with the command: switchport backup interface gigabitEthernet 1/0/12 multicast fast-convergence.
Now, on to VLAN load balancing. It is quite easy; just specify which VLANs you prefer on which links:
Switch(config-if)#switchport backup interface gigabitEthernet1/0/2 prefer vlan 101-200.
If you have VLANs 1-200 on the switch, show interfaces switchport backup will show you:
Vlans Preferred on Active Interface: 1-100
Vlans Preferred on Backup Interface: 101-200
If a link goes down, VLANs that are preferred on that interface will be moved to the other link in the pair. Likewise, when a link returns to service, its preferred VLANs are blocked on the backup and returned to the preferred link.
Be sure to run show interfaces switchport backup detail to see the full status, including link speeds, preemption modes, the MAC address-table move update status.
In summary, the simplicity of Flex Links make it a better choice for carrier and core enterprise networks over the ubiquitous spanning tree protocol. Link-level redundancy is had via STP, but with Flex Links you have more control and better load balancing capabilities. This certainly means that it takes longer to configure since you are planning the layer 2 network manually, but when you need a stable no-surprises link-layer network, Flex Links are definitely the way to go.
2 Comments » Related posts:
- Built-in Security with Cisco IPS
- Cisco AutoQoS: VoIP QoS for Mere Mortals
- Networking 101: Layer 2, Link and Spanning Tree
- Manage Devices and Configurations with Cisco SDM
- What the Heck is a TCAM?
Posted: February 16th, 2010 | Author: charlie | Filed under: Networking | Tags: ccie, ccna, cisco, networks | No Comments »
Let’s talk about TCAM hardware, Cisco SDM templates, and try to answer that elusive question: “why do I have to reboot my router to enable certain features, which in turn disables others?”
First, CAM stands for Content Addressable Memory. A CAM is a special type of memory; some would say the opposite of RAM. With normal computer memory (RAM) the operating system provides an address, and receives the data stored at the supplied address. With a CAM, the operating system supplies the data, and the CAM returns a list of addresses where the data is stored, if it finds any. Furthermore, a CAM searches the entire memory in one operation, so it is considerably faster than RAM.
CAMs are very expensive, so they aren’t normally found in PCs. Even router vendors will sometimes skimp, opting to instead implement advanced software-based searching algorithms to plod through RAM. Most commonly, CAMs and TCAMs are found in network processing devices, including Intel IXP cards and various routers or switches. The most commonly implemented CAMs are called binary CAMs. They search only for ones and zeros; a simple operation. MAC address tables in switches commonly get stored inside binary CAMs. You can bet that any
A Renesas TCAM
switch capable of forwarding Ethernet frames at line-speed gigabit is using CAMs for lookups. If they were using RAM, the operating system would have to remember the address where everything is stored. With CAMs, the operating system can find what it needs in a single operation. In this case desired data is the switchport that data should be sent out, based on the given MAC address, i.e. the essence of a MAC table. Some older Cisco switches running CatOS even opted to call this table the cam table, thereby causing great confusion across the land. Bridge table, forwarding table, mac-address table, cam table; it’s all the same.
Finally, a TCAM is a Ternary CAM. This allows the operating system to match a third state, “X.” The X state is a mask, which means you don’t care what it is. This naturally lends itself to networking, since netmasks operate this way. To calculate a subnet address we mask the bits we don’t care about, and then apply the logical AND operation to the rest. Being able to do this in hardware is a great benefit for routers. Additionally, routers can store their entire routing table in these TCAMs, allowing for very quick lookups. A router with routing tables in TCAMs can find the next-hop destination in a single operation every time instead of trying to search through a tree (or other data structure) in RAM.
Hardware can sometimes seem magic, but it isn’t always transparent. When configuring routers most people will run into a situation where enabling a new feature will require that the Cisco SDM (Switching Database Manager) template be changed. This template is actually a method Cisco uses to assign specific applications to specific TCAM resources.
Some routers will allow you to manually specify how much TCAM space you want to allocate to a specific feature. Others aren’t so nice. They make you choose from a few restrictive templates, which allocate the resources automatically based on a few predetermined settings. For example, on the Cisco 3750, we recently wanted to enable policy-based routing (PBR) to implement a layer 3 jail. The basic idea with template-only routers is that you have to choose where you want most of the optimizations, and compromise on the rest.
For this platform, there are four templates: default, routing, PBR, and VLAN. Each of these tries to allow for a bit more resources allocated to the specified task. For policy routing, we’d have to choose “routing” or “PBR,” which in turn limits the amount of unicast MAC addresses that can be held in TCAMs. Likewise, selecting a VLAN template will make PBR impossible, but allow for more VLAN database information to be held in TCAMs. There are always compromises when we need to use more advanced features. Keeping true with the spirit of router operating systems, there’s also some mysterious side-effects when a new template is chosen. On our specific router, if the PBR template is chosen, the router will become unable to support VPN routing/forwarding tables (VRF). The next unsightly gotcha is that with the IOS version that supports IPv6, you cannot even enable PBR. There is no template to allow both policy routing and IPv6.
Perhaps the main idea of TCAM allocation still isn’t clear. Just because, for example, 8K is allocated to routing tables, this doesn’t mean that you can only have a routing table of that size. There’s always the fallback of process switching. Process switching means that everything will be done by the processor instead of in hardware (TCAMs). Processor intervention is not desirable, mostly because it is much slower than hardware lookups. Also, the processor is supposed to be used for things like sending logs to a syslog server and controlling SSH sessions. If a router doing process switching gets really busy, it may be unable to service your console access attempts. Generally speaking, the more expensive the router, the less it will use the processor.
Hardware is finite, and we always need more. More expensive routers don’t always suffer from the constant struggle for TCAMs because they have enough to support most features that currently exist. Unfortunately, most companies won’t want to purchase the latest and greatest router with seemingly endless hardware resources unless they can justify the added cost by showing a current need for them. So, most of us are stuck having to adjust TCAM allocations.
Further reading: an interesting blog from Plixer.
No Comments yet... be the first » Related posts:
- Networking 101: Subnetting – Slice Up 32-bits
- Networking 101: Layer 2, Link and Spanning Tree
- Networking 101: Understanding Layers
- Are Cisco Flex Links the End of STP?
- Networking 101: More Subnets, and IPv6
Posted: February 14th, 2010 | Author: charlie | Filed under: Networking, Security | Tags: ccna, cisco, networks, Security | No Comments »
Another new feature available in IOS (12.3) is Cisco’s Intrusion Prevention System. An IDS has been part of IOS for a long time, but they recently took it a step further. As part of its Self-Defending Network campaign, Cisco realized that an IPS should be integrated into the network fabric. We’ll explain what this means, and show you how to implement it.
Actively preventing the attack makes it an IPS. The standard old IDS solution means that it can detect and alert, but blocking attacks is not normally part of an IDS’s feature set. Thus, if you want to prevent attacks rather than just receive alerts, you need an IPS. Cisco’s IPS works like any other: you get a signature file, called the Signature Definition File (SDF) by Cisco, and if the IPS finds that a packet matches a signature, it’s blocked.
There are appliances, Catalyst switch modules, and router modules, but IPS is also built-in to certain IOS images now. Since Cisco claims IPS features won’t impact router performance (since the latest release), it may be possible to skip the purchase of a dedicated module for IPS.
The catch, of course, is that an IPS is not robust without constant signature updates. Attacks are constantly evolving, and without an update you aren’t protected against the latest and greatest attacks. Something completely new could sneak in, but the idea is that after the first few attacks Cisco will update the SDF and you’ll be notified that it’s time to download a new version. That’s right, you have to manually download and install a new signature file. This requires a subscription service above and beyond what you pay for SMARTnet. Services for IPS, as it’s called, provides SDF updates and the other features (support, warranty) that SMARTnet does as well. Accordingly, your SMARTnet contract is discounted when you purchase a Cisco Services for IPS contract, according to Cisco’s Q&A documentation.
Configuring IPS for Sensor Modules
There are many different cases for configuring IPS depending on your device. First, we’ll show you how to enable it on any IPS sensor module that uses the IPS 5.1 or later, then we’ll show you how to take advantage of the IOS built-in default IPS features.
The IDS Device Manager (IDM) is a graphical interface for configuring all IDS (and IPS) functionality. If you prefer that, then refer to the Cisco documentation after reading about how it’s done via the CLI here.
The general idea we’re working with here is called the VLAN pair method. This means that we’ll configure two VLANs in a pair group, and all traffic received by a sensor will be inspected and either forwarded on to the other VLAN, or dropped. Up to 255 VLAN pairs can be configured on most sensors.
First we enter configuration mode, then the service interface, and finally select the physical interface that we wish to configure:
sensor#configure terminal
sensor(config)#service interface
sensor(config-int)#physical-interfaces GigabitEthernet0/1
Next, we must configure the VLAN pair (and give it a meaningful description):
sensor(config-int-phy)#subinterface-type inline-vlan-pair
sensor(config-int-phy-inl)#subinterface 1
sensor(config-int-phy-inl-sub)#vlan1 10
sensor(config-int-phy-inl-sub)#vlan2 11
sensor(config-int-phy-inl-sub)#description vlans 10 and 11
Conceptually, the interface will now be added to a virtual sensor, and once it’s enabled it will monitor traffic. We now need to enable a virtual sensor:
sensor(config)#service analysis-engine
sensor(config-ana)#virtual-sensor vs0
Once that’s completed, we simply add the previously-defined subinterface to the sensor, and we’re done:
sensor(config-ana-vir)#physical-interface GigabitEthernet0/2 subinterface-number 1
Configuring IPS for IOS
You can enable IPS features in IOS using the default SDF. Signatures may be added manually to the SDF, or you can pay Cisco for the latest signatures.
First we need to enable what’s called Security Device Event Exchange notifications:
router(config)#ip ips notify sdee
Then we must configure an IPS rule name that will be used for associating with interfaces.
router(config)# ip ips name MYIPSRULES
The next step is to specify where the SDF file will come from. The following command specifies that the file 256MB.sdf can be found in flash memory. You can also specify tftp or any other protocol your Cisco knows how to handle, but it’s best to use flash memory to ensure no dependencies on other servers.
router(config)# ip ips sdf location flash:256MB.sdf
Finally, we simply enable IPS on the interface (in both directions). It is also a good idea to enable IP reassembly on the interface, so that the IPS rule can evaluate entire IP packets at once.
router(config)#interface fastEthernet 0
router(config-if)#ip ips MYIPSRULES in
router(config-if)#ip ips MYIPSRULES out
router(config-if)#ip virtual-reassembly
Now you have a working IPS, based on the file in your flash called 256MB.sdf. That file must be downloaded from Cisco using your CCO login linked to a valid support contract.
The Power of Community
If you don’t feel like paying Cisco for signature updates, you can update the SDF yourself. When a new attack surfaces, you’ll often find Cisco IPS XML signatures posted to various online forums. You can and should use them.
To view your current SDF version, you can run: sh ip ips signatures
To merge the IPS SDF configuration with new information, you can copy in an XML file. Just like copying in any configuration snippet, the updates will be merged, not replaced. Say we got sigs.xml from a helpful network operator. To enable these signatures, we simply run:
router#copy tftp://serer.fqdn/sigs.xml ips-sdf
That’s it! You’ll see that 256MB.sdf on the flash memory is now a bit larger. It’s a good idea (and is recommended by Cisco) to rename 256MB.sdf to avoid confusion, now that you are no longer running a Cisco-sanctioned version.
Enabling IPS on supported routers is quite easy, but can lead to some interesting troubleshooting sessions. Be sure you have a syslog server that your routers all log to: it will save hours of work. Also, search around; you may find a source for XML updates that you wish to trust, and then it’s pretty easy to automate daily merges into your local SDF.
No Comments yet... be the first » Related posts:
- Manage Devices and Configurations with Cisco SDM
- Are Cisco Flex Links the End of STP?
- Cisco AutoQoS: VoIP QoS for Mere Mortals
- Multi-user Security in Linux
- What the Heck is a TCAM?
Posted: February 14th, 2010 | Author: charlie | Filed under: IT Management, Linux / Unix, Networking | Tags: linux, monitoring, networks, NMS | 16 Comments »
Another perfect example of open source software gone commercial is Zenoss. As a full-featured network and service monitoring solution, Zenoss is one of the best monitoring tools available.
Most importantly, Zenoss combines two functionalities. First and foremost an enterprise environment requires host and service monitoring, with notifications. Network monitoring really means checking services, checking that hosts are up (they ping), and possibly writing your own plugins to check various other aspects of a server or network device. Until now, Nagios has filled that role.
Second, once a decent monitoring solution is in place, getting time-based information becomes desirable. Memory and CPU usage is the most prevalent example: if you’re checking available swap space every so often with Nagios, you may know when you start running low. But it may be just as important to see a graph of the last week’s usage. Tools like Cacti or Munin, which collect data frequently and use RRD graphs to display it, are very useful.
Zenoss fills both roles, without the annoying shortcomings prevalent in the alternative solutions. Zenoss uses the terms Availability Monitoring and Performance Monitoring to describe these two fundamental roles.
Performance of monitoring tools is important, and often times overlooked until it becomes a debilitating problem. For example, if you want to chart pretty RRD graphs of systems statistics like available RAM or disk space, Munin is an option. Unfortunately it’s all Perl, and designed in such a way that prevents it from scaling to even moderate amounts of hosts. Cacti is a bit better, but monitoring close to 100 hosts is painful with either option. Along comes Zenoss.
Zenoss is written in Python, and uses a MySQL backend for storage, and by all accounts it appears to perform very well. The really great thing about corporate-backed open source is quality control. The community simply isn’t responsible enough to say, “No, this won’t work, re-implement it.” A company with QA is.
Speaking of features, Zenoss isn’t missing many. Flexibility seems to be top priority–it can monitor hosts with SNMP, Nagios agents, SSH, Windows WMI, and various other mechanisms. Many features they claim are a bit over-inflated, such as ZenPing (marketed as Network Topology Monitoring) but the feature set is rich nonetheless.
Zenoss’s primary functions involve four features:
- Inventory Tracking
- Availability Monitoring
- Performance Monitoring
- Event Monitoring and Management
Inventory tracking claims some sort of “configuration” reporting as well, but it seems very limited. Zenoss will discover your inventory and auto-populate a database. This is great for knowing which IP addresses are in use, for example, but means that “configuration” reporting is limited to an outside observer’s perspective. It can tell you which servers have a Web server running, but it certainly doesn’t deal with the configuration of the Web server. Of course, inventory tracking isn’t limited to automatically discovered information; there are manual input capabilities too.
Availability monitoring is basically Nagios, plus. It can ping, it can monitor Windows machines, and it can pretty much do whatever you need. Even your old Nagios plugins will work with Zenoss. It does generate reports, but much better ones than Nagios is capable of.
Host monitoring, performance monitoring, or whatever you’d like to call it, is quite robust in Zenoss. Some would think it’s light on features, but there’s a good reason that Zenoss requires you use SNMP: it’s much more scalable than SSH’ing to each server every minute. A bit of up-front configuration is required, in that all your hosts will need SNMP configured and working, but it’s completely worth it. Zenoss too uses RRD graphs, and it can generate events and alerts based on pre-defined thresholds.
Finally we come to event monitoring. Zenoss is also encroaching on Splunk‘s territory a bit. It can combine syslog, availability monitoring alerts, SNMP traps, and even Windows event log data. Much like Splunk, Zenoss correlates similar events for easier viewing and troubleshooting. This is the portion that processes all events and generates alerts to pagers or e-mail, taking into account the escalation procedure you’ve defined.
To top it all off, the Zenoss Web interface is top-notch. It includes a customizable “dashboard” for monitoring, and everything is AJAX-enabled. AJAX provides the user experience similar to Splunk and Google’s Gmail.
Marketing fluff aside, Zenoss really does provide a wonderful product. It is, of course, open source and available for free.
At last year’s LISA conference, Zenoss gave a demonstration that sadly coincided with free beer time. Stumbling in toward the end, I demanded one of their free baseball caps, and sat to listen to the last few audience questions. One thing was very obvious: everyone in the room was excited about this product. If hardcore sysadmins are excited, you know this is something worthwhile.
Zenosss is very functional and full of features. It may even be possible to replace three separate pieces of software with this one product: host inventory database, Nagios, and your performance monitoring tool of choice. Maybe even Splunk some day. We can’t wait to see what features they will be adding next.
16 Comments » Related posts:
- Squeeze Your Gigabit NIC for Top Performance
- Managing Virtual Machine and Cloud Sprawl
- Built-in Security with Cisco IPS
- Back to Basics: Unix System Stats Utilities
- Manage Devices and Configurations with Cisco SDM
Posted: February 13th, 2010 | Author: charlie | Filed under: Networking | Tags: linux, networks, NIC, tuning, windows | 1 Comment »
Many new workstations and servers are coming with integrated gigabit network cards nowadays, but quite a few people soon discover that they can’t transfer data much faster than they did with 100 Mb/s network cards. Multiple factors can affect your ability to transfer at higher speeds, and most of them revolve around operating system settings. In this article we will discuss the necessary steps to make your new gigabit enabled server obtain close to gigabit speeds in Linux, FreeBSD, and Windows.
Hardware considerations
First and foremost we must realize that there are hardware limitations to consider. Just because someone throws a gigabit network card in a server doesn’t mean the hardware can keep up. Network cards are normally connected to the PCI bus via a free PCI slot. In older workstation and non server-class motherboards the PCI slots are normally 32 bit, 33MHz. This means they can transfer at speeds of 133MB/s, but since it is a shared bus between many parts of the computer, realistically it’s limited to around 80MB/s in the best case. Gigabit network cards are 1000Mb/s, or 125MB/s. If the PCI bus is only capable of 80MB/s this is a major limiting factor for gigabit network cards. The math works out to 640Mb/s, which is really quite a bit faster than most gigabit network card installations, but remember this is probably the best-case scenario. If there are other hungry data loving PCI cards in the server, you’ll likely see much less throughput. The only solution for overcoming this bottleneck is to purchase a motherboard with a 66MHz PCI slot, which can do 266MB/s. Also, the new 64 bit PCI slots are capable of 532MB/s on a 66MHz bus. These are beginning to come standard on all server-class motherboards.
Assuming we’re using decent hardware that can keep up with the data rates necessary for gigabit, there is now another obstacle – the operating system. For testing, we used two identical servers: Intel Server motherboards, Pentium 4 3.0 GHz, 1GB RAM, integrated 10/100/1000 Intel network card. One was running Gentoo Linux with a 2.6 SMP kernel, and the other is FreeBSD 5.3 with an SMP kernel to take advantage of the Pentium 4’s HyperThreading capabilities. We were lucky to have a gigabit capable switch, but the same results could be accomplished by connecting both servers directly to each other.
Software considerations
For testing speeds between two servers, we don’t want to use FTP or anything that will require data be fetched from disk. Memory to memory transfers are a much better test, and many tools exist to do this. For our tests, we used ttcp (http://www.pcausa.com/Utilities/pcattcp.htm).
The first test between these two servers was not pretty. The maximum rate was around 230 Mb/s, about two times as fast as a 100Mb/s network card. This is an improvement, but far from optimal. In actuality, most people will see even worse performance out of the box. However, with a few minor setting changes, we quickly realized major speed improvements – more than a threefold improvement over the initial test.
Many people recommend setting the MTU of your network interface larger. This basically means telling the network card to send a larger sized Ethernet frame. While this may be useful when connecting two hosts directly together, it becomes less useful when connecting through a switch that doesn’t support larger MTUs. At any rate, this isn’t necessary. 900Mb/s can be attained at the normal 1500 byte MTU setting.
For attaining maximum throughput, the most important options involve TCP window sizes. The TCP window controls the flow of data, and is negotiated during the start of a TCP connection. Using too small of a size will result in slowness, since TCP can only use the smaller of the two end system’s capabilities. It is quite a bit more complex than this, but here’s the information you really need to know:
For both Linux and FreeBSD we’re using the sysctl utility. For all of the following options, entering the command ‘sysctl variable=number’ should do the trick. To view the current settings use: ‘sysctl ’
Maximum window size:
FreeBSD:
kern.ipc.maxsockbuf=262144
Linux:
net.core.wmem_max=8388608
Default window size:
FreeBSD, sending and receiving:
net.inet.tcp.sendspace=65536
net.inet.tcp.recvspace=65536
Linux, sending and receiving:
net.core.wmem_default = 65536
net.core.rmem_default = 65536
RFC 1323:
This enables the useful window scaling options defined in rfc1323, which allows the windows to dynamically get larger than we specified above.
FreeBSD:
net.inet.tcp.rfc1323=1
Linux:
net.ipv4.tcp_window_scaling=1
Buffers:
When sending large amounts of data, we can run the operating system out of buffers. This option should be enabled before attempting to use the above settings. To increase the amount of “mbufs” available:
FreeBSD:
kern.ipc.nmbclusters=32768
Linux:
net.ipv4.tcp_mem= 98304 131072 196608
These quick changes will skyrocket TCP performance. Afterwards we were able to run ttcp and attain around 895 Mb/s every time – quite an impressive data rate. There are other options available for adjusting the UDP datagram sizes as well, but we’re mainly focusing on TCP here.
Windows XP / 2000 Server / Server 2003
The magical location for TCP settings in the registry editor is:
HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Services\Tcpip\Parameters
We need to add a registry DWORD named TcpWindowSize, and enter a sufficiently large size. 131400 (make sure you click on decimal) should be enough.
Tcp1323Opts should be set to 3. This enables both rfc1323 scaling and timestamps.
And similarly to Unix, we also want to increase the TCP buffer sizes:
ForwardBufferMemory 80000
NumForwardPackets 60000
One last important note for Windows XP users needs to be made. If you’re installed service pack 2, then there is another likely culprit of poor network performance. Explained in knowledge base article 842264, Microsoft says that disabling Internet Connection Sharing after an SP2 install should fix performance issues.
The above tweaks should enable your sufficiently fast server to attain much faster data rates over TCP. If your specific application makes significant use of UDP, then it will be worth looking into similar options relating to UDP datagram sizes. Remember, we obtained close to 900Mb/s with a very fast Pentium 4 machine, server-class motherboard, and quality Intel network card. Results may vary wildly, but adjusting the above settings are a necessary step toward realizing your server’s capabilities.
1 Comment » Related posts:
- How Much Server do you Need?
- Zenoss: We Can Ditch Nagios Now
- What the Heck is a TCAM?
- Understanding Linux Virtual Memory
- Back To Basics: Unix Differences in Performing Tasks
