Scalable Traffic Direction and Load Balancing in the "Cloud" with Zeus and Joyent

Last night Zeus and Joyent announced the offering of ZXTM inside Zeus accelerators.

There are some interesting things to note about the offering, the relationship and how the Zeus accelerators work. It also marks the beginning of networked “appliances” being offered at Joyent and a new delivery mechanism and model for Zeus.

Zeus, as you may know, produced one of the better performing web servers and now has the only traffic direction solution that’s offered as software and would be considered “enterprise grade”. As an enterprise grade load-balancer/traffic manager, it’s typically sold and distributed like most enterprise software. ZXTM as offered at Joyent is a new way for software makers like Zeus to ship and sell software, and we’re excited that they trusted us with it. It’s definitely the most accessible and cost-effective way to get ZXTM.

As a fairly regular topic on this blog, we’ve noted that traffic direction is the key to any scalable architecture: HTTP is a stateless protocol (its scale is inherently horizontal) and a domain name requires a relatively small set of IP addresses in the front. There is state then in the IP address and scale requires linking those few front-end IP addresses (e.g. 2-3) with tens to thousands and perhaps back to tens of back-end application nodes. That’s where products like F5’s BIG-IPs and Zeus ZXTM come into play. One is a hardware solution and the other is software.

We’re only offering the full ZXTM, to keep things simple, and everything is basically available within normal accelerator pricing (you bought 2x $500 accelerators before and now they’re a ZXTM cluster). The clustered accelerators provide each customer with an independent and isolated ZXTM setup, complete access to the ZXTM management interface and API, and unprecedented access to network and traffic metrics.

Also a “developer version” of ZXTM will be showing up as standard package in all of our future accelerators. The developer version won’t do all the clustering, in-memory caching, failover and multi-node stuff but Zeus has been kind enough to provide all of us with what is otherwise a feature complete software stack. This will allow anyone to investigate the API, use it on single accelerators and get an idea of the kinds of information and metrics that you can get out of Zeus’s management interface.

For the paid Zeus Accelerators, they’re only available as at least redundant pairs and there are no limitations on the number of backend accelerators or clustering (except the normal practical limits). It took some work for this type of network failover to work within the Joyent cloud (and as far as I know, we’re the only ones that can securely do it). We feel that if you actually need these then they should always be VLAN’ed and clustered, and that is the only way we’re selling them. You can scale your Zeus Accelerators by just adding more memory and CPU power, and their scale is really driven by how much you have to cache in-memory.

With the ability to load-balance and direct gigabits per second now completely in the hands of our customers (and at a moment’s notice), we’re working hard to add more and more unique “functional” accelerators to the mix and in many cases, like ZXTM, they’ll be available as larger, complex and more interesting setups and packages.

So let’s welcome Zeus to the family and Happy Traffic Directing everyone.



Post written by jason