Maintaining App Performance in the Cloud

To read the headlines of late, it would seem that virtualization and cloud alone are the keys to IT nirvana.

But now that the enterprise industry is gaining some practical experience with these new technologies, it’s obvious that things aren’t quite so simple.

For one thing, things like application performance and general data flexibility are still subject to the same old traffic jams that plague traditional infrastructure. The most scalable set of resources in the world won’t be of much use if data can’t get from one place to another.

This is probably why platform vendors of all stripes are falling over themselves to provide the most streamlined I/O architectures through a range of networking, optimization and management techniques.

Naturally, this is fertile ground of startups like Kaminario, which recently unveiled a low-cost DRAM storage appliance designed to ease network bottlenecks for high-throughput applications. The K2 device starts as low as $50,000 for a 500 GB configuration, with the ability to scale up to 12 TB for a cool $1.5 million. It provides 150,000 to 1.5 million IOPS and 1.6 to 16 GBps throughput using the company’s Scale-out Performance Architecture (SPEAR) that provides for independent scalability of processor and storage.

Also building on the appliance model is Dataram, which recently released an upgraded XcelaSAN system that the company says improves IOPS performance 30-fold. Aimed at mid-range Fibre Channel installations, the new Model 100 offers up to 450,000 IOPS along with added features like internal mirrored Flash drives and hot-swappable, redundant power supplies to ensure application and data availability.

In the age of storage/network convergence, however, application performance can suffer greatly without a smooth hand-off between the various protocols, according to Chelsio Communications. This is particularly true for iSCSI, which often has to tap into a host’s general-purpose processor at the expense of both performance and efficiency. The solution, according to Chelsio, is a new network fabric consisting of a T3 adapter that unlocks IP storage traffic to put iSCSI on par with Fibre Channel and InfiniBand.

Brocade, meanwhile, is employing the new IPv6 protocol as a means to enhance its ServerIron ADX application delivery switch. The unit’s latest software release (Ver. 12.3), provides IPv4-to-IPv6 translation, allowing the device to accommodate new IP addresses for mobile, cloud and other resources. At the same time, the switch itself provides for up to three million HTTP connections per second, an increase the company says is needed to handle the expected doubling of Internet bandwidth consumption within the next two years.

In the old days, improving application performance was a relatively simple matter of optimizing the closed environment in which the application existed. In today’s world, that app needs to exhibit peak performance across a wide range of server, storage and network configurations.

This kind of universality is probably out of reach for many organizations, but at least the technology exists to attack the specific weak links in the delivery chain. It will take a bit of work to spot many of these flaws, but the results should be well worth the effort.