The Zeus Blog: Application Delivery, Cloud Computing, Load Balancing, Virtualization

Dynamic scaling with Zeus and the Rackspace Cloud

Wouldn't it be nice to have an infrastructure that expands automatically when your application starts to slow down, and can shrink back when demand is low, guaranteeing both consistent performance and minimized running costs?  Well, with the Zeus Traffic Manager and Rackspace's Cloud you can have just that.

The Zeus Traffic Manager has supported this type of dynamic scaling for some time, utilizing its TrafficScript™ scripting language,  Java™ Extensions, Service Level Monitoring and Control API, but there hasn't been many commercial platforms that could take advantage of the flexibility.  The Rackspace Cloud, however, provides exactly what the traffic manager needs to ensure your applications will always deliver performance as efficiently as possible.

So what are the magic ingredients that the Rackspace Cloud provide, which, when combined with the Zeus Traffic Manager, provide this service delivery Utopia?  Put simply, the ability to provision a new server running a pre-configured software image and on a specified hardware set-up, or to switch one off, using straight-forward API calls.

How do these ingredients fit together then?

• Each request/response that goes through the traffic manager is assigned to a Service Level Monitoring (SLM) class.
• A TrafficScript rule compares the performance data from the SLM class to some pre-determined thresholds.
• If performance is too slow, a Java Extension calls the Rackspace API to provision a new server. When the new server is up and running, the Java Extension then uses the traffic manager's own control API to add the new server into the relevant server pool - performance is boosted.
• If performance is faster than required, a Java Extension uses the traffic manager control API to remove a server from the relevant server pool, and then calls the Rackspace API to delete the server instance - performance reduced, money saved.

If your interested in finding out more about these capabilities in the Rackspace Cloud why not check out the Rackspace Cloud tool-sharing site.

Matt Geldert

Network Performance with VMware Virtualization

You know the sort of opinions and guesses that get circulated, never challenged and eventually become accepted wisdom without ever being proved?   Don’t eat before you swim else you’ll get cramp.  Dress up warm or you’ll catch cold.  Or the myths that take root because their proponents have more voice than their detractors?  A $100 HDMI cable will give you a better picture than a $10 one.

Here’s another one: Network-intensive applications are not suitable for virtualization.  Surely it’s obvious... you take an application (like a software load balancer) that does a lot of network I/O, you put it on top of a virtualization stack, and of course the overhead of virtualization is going to impact performance significantly. 

Zeus has a lot of customers who run our software traffic manager on VMware’s virtualization stack, and they are sometimes curious about performance.  So we ran some tests, using VMware vSphere 4 (aka ESX 4.0):

The tests were a mixture of network-intensive (bandwidth and connections-per-second) and CPU-intensive (SSL transactions per second) workloads.

We compared the performance of a pre-release of ZXTM Virtual Appliance 5.1r2 running on ESX 3.5 and ESX 4.0 to the ZXTM software 5.1r1 running on a typical Linux installation (Ubuntu 8.10).

Compared to ESX 3.5, we found that the network-intensive tests generated approximately 25% better performance on ESX 4.0.  There was little difference in the SSL performance.

The most interesting results come from comparing VMware vSphere 4 (ESX 4.0) with the native performance obtained from the hardware:

The virtual appliance was tested with a number of virtual CPUs; the 4 vCPU result is typically lower than the 3 vCPU result because of contention with the hypervisor processes on CPU0.

For the majority of tests, the performance of the virtual appliance is within 90% of the native system, and in some cases, exceeds native by up to 20%.  The traffic management system runs faster under VMware than natively!

What conclusions can we draw from this?

The myth that workloads that perform a lot of network I/O are not suitable for virtualization has been well any truly debunked, at least when considering VMware virtualization and a sensibly tuned virtual appliance.

Clearly, the VMware hypervisor does not impose a significant performance impediment, and in some cases the combination of VMware and the tuned Virtual Appliance gave better performance than the native installation of ZXTM.

For more information, delve deeper into the test report on www.zeus.com.

Owen Garrett

Software Traffic Management - Guilty As Charged?

When a prospective customer is considering our solution against the hardware appliance brands in the market place one of the objections we encounter is – how can your software-based solution have better performance than that of a hardware appliance solution?

Historically the hardware appliance has been the King in the traffic management space, but it is fair to say that the tide is changing and the concern that many once had over performance of the software-based solution versus the hardware appliance solution is becoming less and less credible.

The case for the defence

In a recent Zeus customer win See Tickets ditched their existing market leading hardware solution and replaced it with the Zeus software solution. See Tickets cited the fact that the appliance solution did not offer the expected performance during peak periods which they needed to ensure their business could remain up and running when it needed to was a huge issue. The software alternative provided by Zeus also offered better reporting which allowed for better-informed business decisions, better ROI, was more flexible and could be deployed rapidly because of the focus on the application rather than the network.  Matt Relf, IT Manager at See Tickets, explains, “Zeus took the time to understand how our application works and the problems we were facing. We now have a solution that delivers significantly better performance and return on investment.” Furthermore, with See Tickets looking to a virtualized environment for the future of their web infrastructure, a software solution seemed the natural choice.

IWOOT, another recent Zeus customer win, decided to ditch their initial hardware traffic management appliances for the Zeus software solution. IWOOT’s business as with See Tickets is reliant on their web infrastructure being highly resilient, scalable and performant during peak periods. IWOOT soon realised that if they wanted to expand and grow their business and make the most of their online operations they needed to move to another solution as their existing solution was not going to cope with increased traffic volumes. Sagar Vadher, Head of IT, IWOOT, explains, “I was not convinced that a software solution would meet our requirements. On a busy day, IWOOT processes 2,000 orders an hour and just a minute of downtime can cost the company dearly. However, I was quickly proven wrong when just 24 hours after the deployment of ZXTM, we were able to remove a bottleneck we had previously found difficult to identify. As a result, we saw an immediate surge in capacity and an increase in uptime to 100%.”

The independent witness

It is not just customers that are seeing the performance capabilities of software based solutions over the hardware appliance. Notable organisations such as Sun Microsystems are now starting to do their own performance testing by taking the software solutions available and placing it on their custom hardware (Sun T5240) and running these against some the hardware appliances. Sun engineers were able to achieve world breaking throughput results and achieved considerable more price per performance than the hardware appliance solutions.

Summing up

So what does this mean to you? Well, if your organization is looking into what options are available in terms of traffic management and the TCO (Total Cost of Ownership), scalability and flexibility of a software option is appealing, don’t be put off my claims that software can’t outperform hardware appliances in terms of performance and even price.

As more organizations learn of the advantages available to them by choosing a software solution to traffic management and application delivery, coupled with organizations looking to virtualization and cloud computing as ways to solve their web infrastructure problems. We are likely to see more examples like the ones in this post.

In my next post I plan to discuss when organizations are most likely to undertake these types of replacement and web infrastructure changes.

How confident are you of your web infrastructure and its performance capabilities?

Mark Gyles

Joyent's MySQL accelerators

The guys at Joyent have recently announced a pre-built, pre-tuned MySQL appliance which offers some very fantastic looking performance benchmarks.  From a technical perspective, Joyent have one of the most complete and capable cloud platforms, and the testaments of their customers back this up (47 billion queries over 13 months on one MySQL appliance, without a reboot or any downtime).

There are a few interesting trends you can draw from this.  Joyent launched their first accelerator application, the Zeus Accelerator in May 2009; with MySQL added to their portfolio and undoubtedly more to follow, this illustrates the growing-up of the Cloud IaaS (Infrastructure-as-a-Service) market and a gradual move up the value stack towards PaaS (Platform-as-a-Service).  Other cloud vendors will certainly follow suit, not forgetting VMware’s purchase of SpringSource which the pundits suggest indicates a move from IaaS to PaaS for them.  Convenience and choice will mark the top IaaS vendors out from the rest who provide little more than a platform to run your virtual machines.

You can also look forward to closer integration between the Zeus  and MySQL accelerators.  Where performance and scalability is key, Zeus’ traffic management technology gives you a way to scale your MySQL capacity horizontally by routing requests transparently based on the permissions of the MySQL user.  This solution is illustrated in the MySQL Scalability knowledgehub article, and shows the huge power of TrafficScript.

Look forward to more innovations from both Joyent and Zeus in the near future.

Owen Garrett