TCP Throughput over Long Fat Networks

Throughput. We all know what that is: how much ‘stuff’ you can move in a period of time, which in the IT world means bits or bytes per second. We also know intuitively about the things which limit throughput; the capacity of a resource and the competing demands for it.

A concept harder to grasp perhaps is the idea that throughput is limited by physical distance and other network delays which is true in the case of TCP. This isn’t something we notice every day because our ubiquitous use of TCP - as HTTP browsing - rarely touches this throughput limit. There can however be real business consequences of this effect anywhere enterprises are moving large volumes of data over long physical distances.

Luckily this TCP throughput limit only exists within individual TCP sessions and doesn’t affect capacity for multiple, concurrent TCP flows. Fortunately there’s also a solution which we’ll come to later.

So what is this TCP throughput constraint and why does it exist?

Back in December 1974 the first TCP specification (RFC675) was published with a 16 bit window size. The window size is the maximum amount of data which can be sent before an acknowledgement is required. The window size can be varied by the receiver to assert flow control but can’t increase beyond 65535 bytes. As the Round Trip Time (RTT) of the network increases, the sender has to wait longer for an acknowledgement before starting to send the next window.

Queuing delays at network hops and long distance propagation delays are significant components of the RTT and certain finite delays can’t be tuned out of the system. Change any of these delay components and the maximum TCP throughput changes and can fluctuate in the course of a transmission and indeed be different in each direction.

By way of example a network path with a RTT of 10ms has a TCP throughput limit of 52Mbit/s including protocol overhead which, with large frames, is about 49Mbit/s of real application data; the so-called ‘goodput’. Halve the RTT and the throughput doubles and vice versa. Even if the Gigabit link to your data centre has an RTT of only 1ms, a bog-standard TCP session could only half fill it. Equally enlightening is that the absolute throughput reduction due to 1ms of extra RTT will halve 500MBit/s to 250Mbit/s but only reduces 50Mbit/s by to 45Mbit/s – the effect of RTT variability is worse at higher throughputs.

The relationship between the path capacity and the path latency determines where performance-limiting factors may lie. Paths with high capacity & high latency will be limited by TCP throughput whereas paths with low capacity and low latency are limited by the data link capacity. This relationship is referred to as the ‘bandwidth delay product’ and is a measure of how much data could be in transit – in the pipe but not yet received - at any point in time. Networks with a high bandwidth delay product are called ‘Long Fat Networks’ or LFNs for short.

If the bandwidth delay product is greater than the TCP window size then the entire window can be in transit. The receiver must sit and wait for it to arrive and then clock it into the buffers before an acknowledgement can be sent. This stop-start pumping effect reduces the effective throughput across LFNs.

So what can be done about this TCP throughput limit?

RFC1323 ‘TCP Extensions for High Performance’ offers a solution called Window Scaling. This is a TCP Option (3) negotiated by both sides during TCP connection establishment to indicate their willingness to shift the advertised window bitwise to the left, doubling it in size each time. A window scale factor of 0 indicates scaling capability but zero scaling, a window scale factor of 1 doubles the maximum advertised window from 65.5KByte to 131KByte. The highest scale factor of 14 can increase the maximum window to just over 1GByte.

Most TCP stacks have incorporated the Window Scale option by default for some time – it’s been around since 1992 after all – but there are a few considerations: Firstly the TCP receive buffer size must be large enough to absorb the increased window size. Secondly there’s a reliability trade-off because losing a single frame within a scaled window will require the whole window to be retransmitted with a further impact on net throughput. Thirdly, TCP congestion avoidance schemes may still limit the ability of the window to ever achieve its maximum size. Fourthly some network devices – most notably firewalls – have been known to re-write the Window Scale option with unpredictable results. The Fifth effect is that aggressive TCP Window scaling can create unfair bandwidth hogging which might not always be desirable.

So how can you engineer the TCP throughput required by particular business applications?

The first step is to understand the real world performance objectives or KPIs. If the raw throughput of individual TCP sessions is paramount then begin by looking at whether the bandwidth-delay product of the proposed network path exceeds the standard TCP window size.

The next step is to make sure that the end systems are tuned both in terms of buffers and two-way support for RFC1323. I’d always recommend throughput testing in conjunction with protocol analysis to validate this and also highlight the effect of congestion avoidance schemes.

I advise impairment testing too because introducing incremental delays and loss under controlled conditions can you get a really good feel for the point at which the tuned system fails to yield throughput returns. A method of monitoring the throughput being achieved may also be necessary prove that the network service is delivering to throughput KPIs and the business is getting what its paying for.

Effective dashboard design. Few and far between.

Everyone has a dashboard don't they? IT service management and monitoring tools all have them. As long as there's a web interface you can customise with traffic lights and gauges then it counts as a dashboard doesn't it? Well no, certainly not according to Stephen Few, the author of beautiful reference books on the art and science of data visualisation.

Few is an evangelist for giving data meaning through visual communication and I'm a recent disciple. In his book 'Information Dashboard Design', he defines a dashboard as the:

'Visual display of the most important information needed to achieve one or more objectives which fits entirely on a single computer screen so it can be monitored at a glance'.

That'll do nicely Stephen.

He critiques examples of dashboards which, whilst you're still uninitiated, don't look too bad. By the time you reach the end of the book you realise that they were truly awful. Poor choice of colours, inappropriate visualisations and non-data pixels are the most common faults. Logos, pie charts and gauges in particular get some stick. Some of BI tool vendors - Business Objects, Oracle, Hyperion, Cognos - are culpable but the greater culprits are the analysts using them to create these monsters.

As well as explaining preattentive processing and the Gestalt principles of visual perception, Few describes the display media which are most useful for dashboards. Summarisation and exception are two key design characteristics; summaries create an aggregate perspective which prompts further questions whilst exceptions, and implicitly the underlying thresholds, draw attention to variations from normals and targets. Customising a dashboard for the audience, their objectives and how they will act on the information is also crucial in deciding on the content. This is where KPI design and visualisation design converge.

I'm often involved in extracting and visualising management information from IT monitoring and service desk tools. These could be the tools which sit there polling or logging day in, day out, but often relegated to an operational notification function. Yes, you can produce traffic lights, charts for WAN or CPU usage and filter huge lists of events but how are you going to use this, er, noise to report the value of the IT service provided to your customers? IT service desks may be better at providing reporting features but suffer from the visual bloat which gets in the way of effective, efficient communication.

Instead of wrestling with the native visualisations that IT management tools provide sometimes it can be useful to re-evaluate what's really important about the information being communicated and work harder on those 'critical few' KPIs and how best to visualise them in a way which helps the audience make better decisions.

I apply Few's techniques everywhere I can. They sit squarely at the intersection of my professional passion for analysing IT performance and a personal interest in aesthetics and creative design.

I highly recommend Few's books if you're working in this area and his website has lots of useful resources. His closing remarks inspire us to design a dashboard which: "makes people's lives better, helps them to work smarter, or gives them what they need to succeed in something that is important to them". As a purpose, there can't be many more worthwhile.

Utilisation. Too busy to manage it?

Utilisation (or utilization if you prefer); In spite of being one of the most widely used terms to describe the performance of IT systems, it often the most misunderstood.  This post examines why this could have real business performance and cost implications.

Utilisation is a common measure of how much of a resource is, or was, being consumed. For storage resources such as physical disks & memory, this is an absolute measure of the available capacity in use eg. 500 Gigabyte of data consumes 50% of a 1 Terabyte disk. This is as intuitive and easy to understand as the glass of water analogy; half full or half empty depending on your outlook.

Misunderstandings can arise though when utilisation is a measure how much of the time a resource was busy. WAN links, CPUs or disk controllers are all resources which, whilst busy processing something, are 100% utilised. Only as the measurement interval increases does the utilisation become less than 100%. A CPU which is 100% busy for 30 seconds and then idle for 30 seconds will be reported as 50% utilised over a 1 minute sampling period.

This might suggest that a time-based utilisation measure is inherently flawed and inaccurate because values reported by monitoring tools are always averaged, even the so-called peaks. In fact this average measure gives rise to something useful because requests which can’t be serviced whilst a resource is busy must wait in a queue for an average time which is determined the average utilisation.

The higher the utilisation of a resource, the higher the probability of encountering queuing and the longer the queues become. Requests are delayed in the queue for the combined service times of the preceding queued requests. So in a system where the throughput of individual components is known, the highest delays – bottlenecks in other words - are encountered wherever the utilisation and service time are highest. Average utilisation then becomes a measure ‘by proxy’ of the queuing delay at each component.

Every part of an application’s response time is bounded by network, CPU & controller queuing delays as well as the actual processing time.  Poor response times mean a poor user experience and reduced productivity. So the time-based utilisation of the end-to-end system resources becomes a critical KPI for managing business performance, IT value perception and capacity costs.

In a converged and virtualised world, there is continual contention for resources so ‘Quality of Service’ scheduling mechanisms which provide queuing control at each bottleneck become a necessary capability. These techniques allow cost efficiency to be maximised, driving resource utilisation as high as possible whilst protecting response times for priority services.

The good news is that utilisation is relatively easy to measure, more so than queuing delays. The real challenge is to collect and present this in a coherent way and translate it into KPIs which reflect real performance impacts. Fail to understand, measure and control utilisation however, and IT could be failing the business.