iHealth now has an API!
Did you ever wish that you had the collective knowlege of the might legions of the F5 Support organization at your beck and call? Maybe you lay in bed dreaming of releasing an army of F5 engineers to scour over your BIG-IPs, looking for potential issues, misconfigurations, and aligning your configurations with known best practices. Taking it even further (we're dreaming, right?) you envisioned a magical robot that would constantly keep watch over your BIG-IPs, checking them for issues, and alerting you when it found something, all without you needing to think about it. On the other hand, maybe you never even thought it was possible and had given up on the dream long ago. F5 has offered the free, web-based iHealth tool (ihealth.f5.com) for years, but perhaps the thought of manually uploading your QKViews, and manually clicking through screens once a week made you throw up your hands, because seriously F5, who has time for that, even with the ability to drag and drop multiple QKViews into the upload page?! Well, we heard rumors of these (not so) secret yearnings for automation, and we're happy to say that we've released an iHealth API to give you the tools you'll need to build your magical automated, proactive BIG-IP monitoring robot. The iHealth REST API provides all kinds of services. Maybe you want to submit a QKView to iHealth to run diagnostics on it. Easy. Oh, actually you want to associate that QKView with an open F5 support case? Easy. You'd like to be able to see that QKView in the iHealth web UI as well? Done. Want to grab the diagnostics output, and run it through your own archiving and reporting system? Easily done. JSON? XML? PDF? Of course! Now that that QKView is in iHealth, do you wish you could get tmsh command output from the moment in time that QKView was taken? Easy. Need a log file from that snapshot to check something? Easy. Your magical robot has the ability (once you teach her the API methods) to manage your QKView collections, retrieve all kinds of things from the QKView itself, get diagnostics output, and much more. UPDATE: QKView Diagnostics documentation can be found here. We encourage you to explore the API methods, and start building your dream robot so that you can spend time on interesting problems, instead of hovering over a web UI, clicking buttons. Get the goods at the iHealth API page.Building a Secure Application DMZ with F5 Distributed Cloud and Equinix Network Edge
Why: Establishing a Secure Application DMZ Enterprises increasingly need to deliver their own applications directly to customers across geographies. Relying solely on external providers for Points of Presence (PoPs) can limit control, visibility, and flexibility. A secure Application Demilitarized Zone (DMZ) empowers organizations to: Establish their own PoPs for internet-facing applications. Maintain control over security, compliance, and performance. Deliver applications consistently across regions. Reduce dependency on third-party infrastructure. This approach enables enterprises to build a globally distributed application delivery footprint tailored to their business needs. What: A Unified Solution to Secure Global Application Delivery The joint solution integrates F5 Distributed Cloud (F5XC) Customer Edge (CE) deployed via the Equinix Network Edge Marketplace, with Equinix Fabric to create a strategic point of control for secure, scalable application delivery. Key Capabilities Secure Ingress/Egress: CE devices serve as secure gateways for public-facing applications, integrating WAF, API protection, and DDoS mitigation. Global Reach: Equinix’s infrastructure enables CE deployment in strategic locations worldwide. Multi cloud Networking: Seamless connectivity across public clouds, private data centers, and edge locations. Centralized Management: F5XC Console provides unified visibility, policy enforcement, and automation. Together, these components form a cohesive solution that supports enterprise-grade application delivery with security, performance, and control. How: Architectural Overview Core Components F5XC Customer Edge (CE): Deployed as a virtual network function at Equinix PoPs, CE serves as the secure entry point for applications. F5 Distributed Cloud Console: Centralized control plane for managing CE devices, policies, and analytics. Equinix Network Edge Marketplace: Enables rapid provisioning of CE devices as virtual appliances. Equinix Fabric: High-performance interconnectivity between CE devices, clouds, and data centers. Key Tenets of the Solution Strategic Point of Control - CE becomes the enterprise’s own PoP, enabling secure and scalable delivery of applications. Unified Security Posture - Integrated WAF, API security, and DDoS protection across all CE locations. Consistent Policy Enforcement - Centralized control plane ensures uniform security and compliance policies. Multicloud and Edge Flexibility - Seamless connectivity across AWS, Azure, GCP, private clouds, and data centers. Rapid Deployment - CE provisioning via Equinix Marketplace reduces time-to-market and operational overhead. Partner and Customer Connectivity - Supports business partner exchanges and direct customer access without traditional networking complexity. Additional Links Multicloud chaos ends at the Equinix Edge with F5 Distributed Cloud CE F5 and Equinix Partnership Equinix Fabric Overview Secure Extranet with Equinix Fabric and F5 Distributed Cloud Additional Equinix and F5 partner informationBIG-IP Next Edge Firewall CNF for Edge workloads
Introduction The CNF architecture aligns with cloud-native principles by enabling horizontal scaling, ensuring that applications can expand seamlessly without compromising performance. It preserves the deterministic reliability essential for telecom environments, balancing scalability with the stringent demands of real-time processing. More background information about what value CNF brings to the environment, https://community.f5.com/kb/technicalarticles/from-virtual-to-cloud-native-infrastructure-evolution/342364 Telecom service providers make use of CNFs for performance optimization, Enable efficient and secure processing of N6-LAN traffic at the edge to meet the stringent requirements of 5G networks. Optimize AI-RAN deployments with dynamic scaling and enhanced security, ensuring that AI workloads are processed efficiently and securely at the edge, improving overall network performance. Deploy advanced AI applications at the edge with the confidence of carrier-grade security and traffic management, ensuring real-time processing and analytics for a variety of edge use cases. CNF Firewall Implementation Overview Let’s start with understanding how different CRs are enabled within a CNF implementation this allows CNF to achieve more optimized performance, Capex and Opex. The traditional way of inserting services to the Kubernetes is as below, Moving to a consolidated Dataplane approach saved 60% of the Kubernetes environment’s performance The F5BigFwPolicy Custom Resource (CR) applies industry-standard firewall rules to the Traffic Management Microkernel (TMM), ensuring that only connections initiated by trusted clients will be accepted. When a new F5BigFwPolicy CR configuration is applied, the firewall rules are first sent to the Application Firewall Management (AFM) Pod, where they are compiled into Binary Large Objects (BLOBs) to enhance processing performance. Once the firewall BLOB is compiled, it is sent to the TMM Proxy Pod, which begins inspecting and filtering network packets based on the defined rules. Enabling AFM within BIG-IP Controller Let’s explore how we can enable and configure CNF Firewall. Below is an overview of the steps needed to set up the environment up until the CNF CRs installations [Enabling the AFM] Enabling AFM CR within BIG-IP Controller definition global: afm: enabled: true pccd: enabled: true f5-afm: enabled: true cert-orchestrator: enabled: true afm: pccd: enabled: true image: repository: "local.registry.com" [Configuration] Example for Firewall policy settings apiVersion: "k8s.f5net.com/v1" kind: F5BigFwPolicy metadata: name: "cnf-fw-policy" namespace: "cnf-gateway" spec: rule: - name: allow-10-20-http action: "accept" logging: true servicePolicy: "service-policy1" ipProtocol: tcp source: addresses: - "2002::10:20:0:0/96" zones: - "zone1" - "zone2" destination: ports: - "80" zones: - "zone3" - "zone4" - name: allow-10-30-ftp action: "accept" logging: true ipProtocol: tcp source: addresses: - "2002::10:30:0:0/96" zones: - "zone1" - "zone2" destination: ports: - "20" - "21" zones: - "zone3" - "zone4" - name: allow-us-traffic action: "accept" logging: true source: geos: - "US:California" destination: geos: - "MX:Baja California" - "MX:Chihuahua" - name: drop-all action: "drop" logging: true ipProtocol: any source: addresses: - "::0/0" - "0.0.0.0/0" [Logging & Monitoring] CNF firewall settings allow not only local logging but also to use HSL logging to external logging destinations. apiVersion: "k8s.f5net.com/v1" kind: F5BigLogProfile metadata: name: "cnf-log-profile" namespace: "cnf-gateway" spec: name: "cnf-logs" firewall: enabled: true network: publisher: "cnf-hsl-pub" events: aclMatchAccept: true aclMatchDrop: true tcpEvents: true translationFields: true Verifying the CNF firewall settings can be done through the sidecar container kubectl exec -it deploy/f5-tmm -c debug -n cnf-gateway – bash tmctl -d blade fw_rule_stat context_type context_name ------------ ------------------------------------------ virtual cnf-gateway-cnf-fw-policy-SecureContext_vs rule_name micro_rules counter last_hit_time action ------------------------------------ ----------- ------- ------------- ------ allow-10-20-http-firewallpolicyrule 1 2 1638572860 2 allow-10-30-ftp-firewallpolicyrule 1 5 1638573270 2 Conclusion To conclude our article, we showed how CNFs with consolidated data planes help with optimizing CNF deployments. In this article we went through the overview of BIG-IP Next Edge Firewall CNF implementation, sample configuration and monitoring capabilities. More use cases to cover different use cases to be following. Related content F5BigFwPolicy BIG-IP Next Cloud-Native Network Functions (CNFs) CNF HomeUsing n8n To Orchestrate Multiple Agents
I’ve been heads-down building a series of AI step-by-step labs, and this one might be my favorite so far: a practical, cost-savvy “mixture of experts” architectural pattern you can run with n8n and self-hosted models on Ollama. The idea is simple. Not every prompt needs a heavyweight reasoning model. In fact, most don’t. So we put a small, fast model in front to classify the user’s request—coding, reasoning, or something else—and then hand that prompt to the right expert. That way, you keep your spend and latency down, and only bring out the big guns when you really need them. Architecture at a glance: Two hosts: one for your models (Ollama) and one for your n8n app. Keeping these separate helps n8n stay snappy while the model server does the heavy lifting. Docker everywhere, with persistent volumes for both Ollama and n8n so nothing gets lost across restarts. Optional but recommended: NVIDIA GPU on the model host, configured with the NVIDIA Container Toolkit to get the most out of inference. On the model server, we spin up Ollama and pull a small set of targeted models: deepseek-r1:1.5b for the classifier and general chit-chat deepseek-r1:7b for the reasoning agent (this is your “brains-on” model) codellama:latest for coding tasks (Python, JSON, Node.js, iRules, etc.) llama3.2:3b as an alternative generalist On the app server, we run n8n. Inside n8n, the flow starts with the “On Chat Message” trigger. I like to immediately send a test prompt so there’s data available in the node inspector as I build. It makes mapping inputs easier and speeds up debugging. Next up is the Text Classifier node. The trick here is a tight system, prompt and clear categories: Categories: Reasoning and Coding Options: When no clear match → Send to an “Other” branch Optional: You can allow multiple matches if you want the same prompt to hit more than one expert. I’ve tried both approaches. For certain, ambiguous asks, allowing multiple can yield surprisingly strong results. I attach deepseek-r1:1.5b to the classifier. It’s inexpensive and fast, which is exactly what you want for routing. In the System Prompt Template, I tell it: If a prompt explicitly asks for coding help, classify it as Coding If it explicitly asks for reasoning help, classify it as Reasoning Otherwise, pass the original chat input to a Generalist From there, each classifier output connects to its own AI Agent node: Reasoning Agent → deepseek-r1:7b Coding Agent → codellama:latest Generalist Agent (the “Other” branch) → deepseek-r1:1.5b or llama3.2:3b I enable “Retry on Fail” on the classifier and each agent. In my environment (cloud and long-lived connections), a few retries smooth out transient hiccups. It’s not a silver bullet, but it prevents a lot of unnecessary red Xs while you’re iterating. Does this actually save money? If you’re paying per token on hosted models, absolutely. You’re deferring the expensive reasoning calls until a small model decides they’re justified. Even self-hosted, you’ll feel the difference in throughput and latency. CodeLlama crushes most code-related queries without dragging a reasoning model into it. And for general questions—“How do I make this sandwich?”—A small generalist is plenty. A few practical notes from the build: Good inputs help. If you know you’re asking for code, say so. Your classifier and downstream agent will have an easier time. Tuning beats guessing. Spend time on the classifier’s system prompt. Small changes go a long way. Non-determinism is real. You’ll see variance run-to-run. Between retries, better prompts, and a firm “When no clear match” path, you can keep that variance sane. Bigger models, better answers. If you have the budget or hardware, plugging in something like Claude, GPT, or a higher-parameter DeepSeek will lift quality. The routing pattern stays the same. Where to take it next: Wire this to Slack so an engineering channel can drop prompts and get routed answers in place. Add more “experts” (e.g., a data-analysis agent or an internal knowledge agent) and expand your classifier categories. Log token counts/latency per branch so you can actually measure savings and adjust thresholds/models over time. This is a lab, not a production, but the pattern is production-worthy with the right guardrails. Start small, measure, tune, and only scale up the heavy models where you’re seeing real business value. Let me know what you build—especially if you try multi-class routing and send prompts to more than one expert. Some of the combined answers I’ve seen are pretty great. Here's the lab in our git, if you'd like to try it out for yourself. If video is more your thing, try this: Thanks for building along, and I’ll see you in the next lab.
Extend visibility - BIG-IP joins forces with CrowdStrike
Introduction The traditional focus in cybersecurity has prioritized endpoints like laptops and mobiles with EDR, as they are key entry points for intrusions. Modern threats target the full network infrastructure, like routers, ADCs, firewalls, servers, VMs, and cloud instances, as interconnected endpoints. All network software is a potential target in today’s sprawling attack surface. Summarizing some of those blind points below, Servers, including hardware, VMs, and cloud instances: Often under-monitored, rapid spin-up creates ephemeral risks for exfiltration and lateral movement. Network appliances: Enable traffic redirection, data sniffing, or backdoors, if compromised. Application delivery components: Vulnerable to session hijacking, code injection, or DDoS, due to high-traffic processing. Falcon sensor integration In this section, we go through download and installation steps, and observe how the solution works with detecting/blocking malicious packages. For more information, follow our KB articles, https://my.f5.com/manage/s/article/K000157015 Related content K000157015: Getting Started with Falcon sensor for BIG-IP K000156881: Install Falcon sensor for BIG-IP on the BIG-IP system K000157014: F5 Support for Falcon for BIG-IP https://www.f5.com/partners/technology-alliances/crowdstrike
How to get a F5 BIG-IP VE Developer Lab License
(applies to BIG-IP TMOS Edition) To assist operational teams teams improve their development for the BIG-IP platform, F5 offers a low cost developer lab license. This license can be purchased from your authorized F5 vendor. If you do not have an F5 vendor, and you are in either Canada or the US you can purchase a lab license online: CDW BIG-IP Virtual Edition Lab License CDW Canada BIG-IP Virtual Edition Lab License Once completed, the order is sent to F5 for fulfillment and your license will be delivered shortly after via e-mail. F5 is investigating ways to improve this process. To download the BIG-IP Virtual Edition, log into my.f5.com (separate login from DevCentral), navigate down to the Downloads card under the Support Resources section of the page. Select BIG-IP from the product group family and then the current version of BIG-IP. You will be presented with a list of options, at the bottom, select the Virtual-Edition option that has the following descriptions: For VMware Fusion or Workstation or ESX/i: Image fileset for VMware ESX/i Server For Microsoft HyperV: Image fileset for Microsoft Hyper-V KVM RHEL/CentoOS: Image file set for KVM Red Hat Enterprise Linux/CentOS Note: There are also 1 Slot versions of the above images where a 2nd boot partition is not needed for in-place upgrades. These images include _1SLOT- to the image name instead of ALL. The below guides will help get you started with F5 BIG-IP Virtual Edition to develop for VMWare Fusion, AWS, Azure, VMware, or Microsoft Hyper-V. These guides follow standard practices for installing in production environments and performance recommendations change based on lower use/non-critical needs for development or lab environments. Similar to driving a tank, use your best judgement. Deploying F5 BIG-IP Virtual Edition on VMware Fusion Deploying F5 BIG-IP in Microsoft Azure for Developers Deploying F5 BIG-IP in AWS for Developers Deploying F5 BIG-IP in Windows Server Hyper-V for Developers Deploying F5 BIG-IP in VMware vCloud Director and ESX for Developers Note: F5 Support maintains authoritative Azure, AWS, Hyper-V, and ESX/vCloud installation documentation. VMware Fusion is not an official F5-supported hypervisor so DevCentral publishes the Fusion guide with the help of our Field Systems Engineering teams.BIG-IP Configuration Conversion Scripts
Kirk Bauer, John Alam, and Pete White created a handful of perl and/or python scripts aimed at easing your migration from some of the “other guys” to BIG-IP. While they aren’t going to map every nook and cranny of the configurations to a BIG-IP feature, they will get you well along the way, taking out as much of the human error element as possible. Links to the codeshare articles below. Cisco ACE (perl) Cisco ACE via tmsh (perl) Cisco ACE (python) Cisco CSS (perl) Cisco CSS via tmsh (perl) Cisco CSM (perl) Citrix Netscaler (perl) Radware via tmsh (perl) Radware (python)
A Brief Introduction To External Application Verification Monitors
Background EAVs (External Application Verification) monitors are one of most useful and extensible features of the BIG-IP product line. They give the end user the ability to utilize the underlying Linux operating system to perform complex and thorough service checks. Given a service that does not have a monitor provided, a lot of users will assign the closest related monitor and consider the solution complete. There are more than a few cases where a TCP or UDP monitor will mark a service “up” even while the service is unresponsive. EAVs give us the ability to dive much deeper than merely performing a 3-way handshake and neglecting the other layers of the application or service. How EAVs Work An EAV monitor is an executable script located on the BIG-IP’s file system (usually under /usr/bin/monitors) that is executed at regular intervals by the bigd daemon and reports its status. One of the most common misconceptions (especially amongst those with *nix backgrounds) is that the exit status of the script dictates the fate of the pool member. The exit status has nothing to do with how bigd interprets the pool member’s health. Any output to stdout (standard output) from the script will mark the pool member “up”. This is a nuance that should receive special attention when architecting your next EAV. Analyze each line of your script and make sure nothing will inadvertently get directed to stdout during monitor execution. The most common example is when someone writes a script that echoes “up” when the checks execute correctly and “down” when they fail. The pool member will be enabled by the BIG-IP under both circumstances rendering a useless monitor. Bigd automatically provides two arguments to the EAV’s script upon execution: node IP address and node port number. The node IP address is provided with an IPv6 prefix that may need to be removed in order for the script to function correctly. You’ll notice we remove the “::ffff://” prefix with a sed substitution in the example below. Other arguments can be provided to the script when configured in the UI (or command line). The user-provided arguments will have offsets of $3, $4, etc. Without further ado, let’s take a look at a service-specific monitor that gives us a more complete view of the application’s health. An Example I have seen on more than one occasion where a DNS pool member has successfully passed the TCP monitor, but the DNS service was unresponsive. As a result, a more invasive inspection is required to make sure that the DNS service is in fact serving valid responses. Let’s take a look at an example: #!/bin/bash # $1 = node IP # $2 = node port # $3 = hostname to resolve [[ $# != 3 ]] && logger -p local0.error -t ${0##*/} -- "usage: ${0##*/} <node IP> <node port> <hostname to resolve>" && exit 1 node_ip=$(echo $1 | sed 's/::ffff://') dig +short @$node_ip $3 IN A &> /dev/null [[ $? == 0 ]] && echo “UP” We are using the dig (Domain Information Groper) command to query our DNS server for an A record. We use the exit status from dig to determine if the monitor will pass. Notice how the script will never output anything to stdout other than “UP” in the case of success. If there aren’t enough arguments for the script to proceed, we output the usage to /var/log/ltm and exit. This is a very simple 13 line script, but effective example. The Takeaways The command should be as lightweight and efficient as possible If the same result can be accomplished with a built-in monitor, use it EAV monitors don’t rely on the command’s exit status, only standard output Send all error and informational messages to logger instead of stdout or stderr (standard error) “UP” has no significance, it is just a series of character sent to stdout, the monitor would still pass if the script echoed “DOWN” Conclusion When I first discovered EAV monitors, it opened up a whole realm of possibilities that I could not accomplish with built in monitors. It gives you the ability to do more thorough checking as well as place logic in your monitors. While my example was a simple bash script, BIG-IP also ships with Perl and Python along with their standard libraries, which offer endless possibilities. In addition to using the built-in commands and libraries, it would be just as easy to write a monitor in a compiled language (C, C++, or whatever your flavor may be) and statically compile it before uploading it to the BIG-IP. If you are new to EAVs, I hope this gives you the tools to make your environments more robust and resilient. If you’re more of a seasoned veteran, we’ll have more fun examples in the near future.F5 Predicts: Education gets personal
The topic of education is taking centre stage today like never before. I think we can all agree that education has come a long way from the days where students and teachers were confined to a classroom with a chalkboard. Technology now underpins virtually every sector and education is no exception. The Internet is now the principal enabling mechanism by which students assemble, spread ideas and sow economic opportunities. Education data has become a hot topic in a quest to transform the manner in which students learn. According to Steven Ross, a professor at the Centre for Research and Reform in Education at Johns Hopkins University, the use of data to customise education for students will be the key driver for learning in the future[1]. This technological revolution has resulted in a surge of online learning courses accessible to anyone with a smart device. A two-year assessment of the massive open online courses (MOOCs) created by HarvardX and MITx revealed that there were 1.7 million course entries in the 68 MOOC [2]. This translates to about 1 million unique participants, who on average engage with 1.7 courses each. This equity of education is undoubtedly providing vast opportunities for students around the globe and improving their access to education. With more than half a million apps to choose from on different platforms such as the iOS and Android, both teachers and students can obtain digital resources on any subject. As education progresses in the digital era, here are some considerations for educational institutions to consider: Scale and security The emergence of a smogasborad of MOOC providers, such as Coursera and edX, have challenged the traditional, geographical and technological boundaries of education today. Digital learning will continue to grow driving the demand for seamless and user friendly learning environments. In addition, technological advancements in education offers new opportunities for government and enterprises. It will be most effective if provided these organisations have the ability to rapidly scale and adapt to an all new digital world – having information services easily available, accessible and secured. Many educational institutions have just as many users as those in large multinational corporations and are faced with the issue of scale when delivering applications. The aim now is no longer about how to get fast connection for students, but how quickly content can be provisioned and served and how seamless the user experience can be. No longer can traditional methods provide our customers with the horizontal scaling needed. They require an intelligent and flexible framework to deploy and manage applications and resources. Hence, having an application-centric infrastructure in place to accelerate the roll-out of curriculum to its user base, is critical in addition to securing user access and traffic in the overall environment. Ensuring connectivity We live in a Gen-Y world that demands a high level of convenience and speed from practically everyone and anything. This demand for convenience has brought about reform and revolutionised the way education is delivered to students. Furthermore, the Internet of things (IoT), has introduced a whole new raft of ways in which teachers can educate their students. Whether teaching and learning is via connected devices such as a Smart Board or iPad, seamless access to data and content have never been more pertinent than now. With the increasing reliance on Internet bandwidth, textbooks are no longer the primary means of educating, given that students are becoming more web oriented. The shift helps educational institutes to better personalise the curriculum based on data garnered from students and their work. Duty of care As the cloud continues to test and transform the realms of education around the world, educational institutions are opting for a centralised services model, where they can easily select the services they want delivered to students to enhance their learning experience. Hence, educational institutions have a duty of care around the type of content accessed and how it is obtained by students. They can enforce acceptable use policies by only delivering content that is useful to the curriculum, with strong user identification and access policies in place. By securing the app, malware and viruses can be mitigated from the institute’s environment. From an outbound perspective, educators can be assured that students are only getting the content they are meant to get access to. F5 has the answer BIG-IP LTM acts as the bedrock for educational organisations to provision, optimise and deliver its services. It provides the ability to publish applications out to the Internet in a quickly and timely manner within a controlled and secured environment. F5 crucially provides both the performance and the horizontal scaling required to meet the highest levels of throughput. At the same time, BIG-IP APM provides schools with the ability to leverage virtual desktop infrastructure (VDI) applications downstream, scale up and down and not have to install costly VDI gateways on site, whilst centralising the security decisions that come with it. As part of this, custom iApps can be developed to rapidly and consistently deliver, as well as reconfigure the applications that are published out to the Internet in a secure, seamless and manageable way. BIG-IP Application Security Manager (ASM) provides an application layer security to protect vital educational assets, as well as the applications and content being continuously published. ASM allows educational institutes to tailor security profiles that fit like a glove to wrap seamlessly around every application. It also gives a level of assurance that all applications are delivered in a secure manner. Education tomorrow It is hard not to feel the profound impact that technology has on education. Technology in the digital era has created a new level of personalised learning. The time is ripe for the digitisation of education, but the integrity of the process demands the presence of technology being at the forefront, so as to ensure the security, scalability and delivery of content and data. The equity of education that technology offers, helps with addressing factors such as access to education, language, affordability, distance, and equality. Furthermore, it eliminates geographical boundaries by enabling the mass delivery of quality education with the right policies in place. [1] http://www.wsj.com/articles/SB10001424052702304756104579451241225610478 [2] http://papers.ssrn.com/sol3/papers.cfm?abstract_id=2586847Is TCP's Nagle Algorithm Right for Me?
Of all the settings in the TCP profile, the Nagle algorithm may get the most questions. Designed to avoid sending small packets wherever possible, the question of whether it's right for your application rarely has an easy, standard answer. What does Nagle do? Without the Nagle algorithm, in some circumstances TCP might send tiny packets. In the case of BIG-IP®, this would usually happen because the server delivers packets that are small relative to the clientside Maximum Transmission Unit (MTU). If Nagle is disabled, BIG-IP will simply send them, even though waiting for a few milliseconds would allow TCP to aggregate data into larger packets. The result can be pernicious. Every TCP/IP packet has at least 40 bytes of header overhead, and in most cases 52 bytes. If payloads are small enough, most of the your network traffic will be overhead and reduce the effective throughput of your connection. Second, clients with battery limitations really don't appreciate turning on their radios to send and receive packets more frequently than necessary. Lastly, some routers in the field give preferential treatment to smaller packets. If your data has a series of differently-sized packets, and the misfortune to encounter one of these routers, it will experience severe packet reordering, which can trigger unnecessary retransmissions and severely degrade performance. Specified in RFC 896 all the way back in 1984, the Nagle algorithm gets around this problem by holding sub-MTU-sized data until the receiver has acked all outstanding data. In most cases, the next chunk of data is coming up right behind, and the delay is minimal. What are the Drawbacks? The benefits of aggregating data in fewer packets are pretty intuitive. But under certain circumstances, Nagle can cause problems: In a proxy like BIG-IP, rewriting arriving packets in memory into a different, larger, spot in memory taxes the CPU more than simply passing payloads through without modification. If an application is "chatty," with message traffic passing back and forth, the added delay could add up to a lot of time. For example, imagine a network has a 1500 Byte MTU and the application needs a reply from the client after each 2000 Byte message. In the figure at right, the left diagram shows the exchange without Nagle. BIG-IP sends all the data in one shot, and the reply comes in one round trip, allowing it to deliver four messages in four round trips. On the right is the same exchange with Nagle enabled. Nagle withholds the 500 byte packet until the client acks the 1500 byte packet, meaning it takes two round trips to get the reply that allows the application to proceed. Thus sending four messages takes eight round trips. This scenario is a somewhat contrived worst case, but if your application is more like this than not, then Nagle is poor choice. If the client is using delayed acks (RFC 1122), it might not send an acknowledgment until up to 500ms after receipt of the packet. That's time BIG-IP is holding your data, waiting for acknowledgment. This multiplies the effect on chatty applications described above. F5 Has Improved on Nagle The drawbacks described above sound really scary, but I don't want to talk you out of using Nagle at all. The benefits are real, particularly if your application servers deliver data in small pieces and the application isn't very chatty. More importantly, F5® has made a number of enhancements that remove a lot of the pain while keeping the gain: Nagle-aware HTTP Profiles: all TMOS HTTP profiles send a special control message to TCP when they have no more data to send. This tells TCP to send what it has without waiting for more data to fill out a packet. Autonagle: in TMOS v12.0, users can configure Nagle as "autotuned" instead of simply enabling or disabling it in their TCP profile. This mechanism starts out not executing the Nagle algorithm, but uses heuristics to test if the receiver is using delayed acknowledgments on a connection; if not, it applies Nagle for the remainder of the connection. If delayed acks are in use, TCP will not wait to send packets but will still try to concatenate small packets into MSS-size packets when all are available. [UPDATE: v13.0 substantially improves this feature.] One small packet allowed per RTT: beginning with TMOS® v12.0, when in 'auto' mode that has enabled Nagle, TCP will allow one unacknowledged undersize packet at a time, rather than zero. This speeds up sending the sub-MTU tail of any message while not allowing a continuous stream of undersized packets. This averts the nightmare scenario above completely. Given these improvements, the Nagle algorithm is suitable for a wide variety of applications and environments. It's worth looking at both your applications and the behavior of your servers to see if Nagle is right for you.
