For IT professionals, one of the real world implications of Moore’s law – that the number of transistors in a dense integrated circuit doubles approximately every two years – was that CPU power doubled roughly every two years also. Around the 2006 timeframe, however, the dramatic increases in CPU speed that we’d long taken for granted began to plateau. This meant that for servers to become more powerful from a processing point of view, they would need to scale in other ways. Manufacturers responded by putting multiple CPUs in a system, which then gave way to having multiple processor cores on a single CPU. A few years later in the late 2000s, when eight-core CPUs and ultra-high-speed solid state drives (SSDs) became affordable and readily available – making dramatic increases in CPU the norm once again – the performance dynamic within many data centers began to shift.
IT performance has long been predicated on a three-legged stool of 1) processing power, 2) storage capacity, and 3) network throughput. And for most of the past fifteen years or so, the hardware wars were pretty much exclusively fought out on processing and storage terrain. This was because when gigabit Ethernet network interface controllers (NICs) became commercially available around 1999, they had dramatically more capacity to handle traffic produced by the processors of the day than the Fast Ethernet technologies they replaced. It’s taken more than a decade for processing power and storage speed/capacity to catch up, but within the commercial data centers where much of the world’s email marketing and e-business computing gets handled, it’s clear that single-gigabit Ethernet will no longer suffice to get the full performance out of today’s multi-core servers.
The Need for Speed
Over a decade ago, Momentum set the standard as the best performing (fastest delivery speeds, highest throughput) email platform on the market. When Momentum was first designed, a state-of-the-art hardware server had one or maybe two CPUs. But, as we’ve seen, things have changed in the hardware space, and so too has Momentum changed. As we announced earlier this year, Momentum has been re-architected to allow customers to fully utilize the resources of today’s multi-core server hardware – with the ability to run multiple event loops per CPU. Unlike most commercially available MTAs that operate on a single master event loop, Momentum now has the ability to leverage virtually all available I/O, CPU, and memory resources on modern systems that offer multiple cores and extremely fast I/Os.
The net result with Supercharger is that we’ve seen significant performance boosts on all systems that had available system resources (disk, I/O and CPU). The amount of boost depends on the available resources on the server in question. A 4x gain is a good baseline; or performance increases could be significantly higher for a system with plenty of available CPU and incredibly fast disks. During tests run within a customer’s production environment, Momentum with Supercharger was able to send in excess of 18 million messages per hour. This rate, however, saturated the gigabit Ethernet interface on the system, indicating that further performance gains would have still been possible with a properly provisioned network. Indicating also that the main roadblock to performance within high-volume messaging implementations going forward is likely to be the network interface. Welcome to the post single-gigabit age.
Learn more about unique features that set our email infrastructure apart from commodity MTAs.