FLASH, AH-AAAH! Saviour of the universe Rackspace cloud?
1.2TB SSD army to storm Jeff's Virtualization Palace
The "fanatical service" acolytes of Rackspace have added a hefty flash-backed range of servers to their cloud as part of a strategic revamp.
The upgrade of the company's cloud platform was announced on Tuesday and sees the Texan data wrangler press some high-memory "performance cloud servers" into duty for its battle with Amazon.
The new servers come with RAID-10 SSD drives, Intel Xeon E5 processors, up to 120GB of RAM, and 40Gbps of networking bandwidth.
With the new system, Rackspace is acknowledging that it won't be able to stay alive in the cloud market on the benefit of "fanatical service" alone and instead needs to devote some major investment to the underlying infrastructure that lives in its data center.
Prices for the new servers range from $0.04 per hour for 1GB of RAM, 20GB of SSD storage, 1 virtual CPU, and a guaranteed 200Mbps of network bandwidth.
This goes all the way up to a whopping configuration with 120GB of RAM, 1.2TB of flash, 32 vCPUs, and 10,000Mbps of bandwidth – though this gigantic system comes with a price tag to match, costing developers some $5.44 per hour.
The most powerful Rackspace servers offer a RAM config that dwarfs instances available in Google Compute Engine (which tops out at 52GB) and Windows Azure (56GB) though it comes behind Amazon's gigantic 244GB "high memory cluster eight extra large" instance (at $3.50 per hour).
However, as far as we can tell from our evaluations, no other consumer cloud service will give you a system that pairs this much RAM with 1.2TB of attached flash. But storage is useless unless you have the bandwidth to shuttle data in and out of it, so Rackspace has upgraded its networking systems as well, CTO John Engates said.
Each node now has "multiple, bonded 10GbE connections" according to Engates, and splits traffic per node between public internet traffic, and block storage and associated services traffic.
"You're not going to have any contention between storage traffic to our cloud block storage," he told El Reg.
"The frontend 20GbE would be for connections to the internet, connections to your other instances, maybe to cloud files - public object storage," he added. "They also are redundant so that if for some reason one of those 20 gig bonded connections can have problems they can each fail over to one another."
Rackspace has been motivated to bring out the new infrastructure in response to the relentless rise of Amazon Web Services, and the increasingly well-armed clouds fielded by companies such as Microsoft and Google, which outspend the company on data center infrastructure by ten or twenty times.
The performance servers will be sold alongside the standard servers at launch, but over time the company will phase out the older offerings.
"We're doing this because we don't want to break any code, partners, etc. that may be assuming standard," Rackspace director of product strategy Eric Carlin told El Reg via email.
"Because performance servers are more performance for a better price, we expect performance to become the default choice for new and existing customers. As quickly as possible, we will begin end of sale of standard for new customers. We will then phase out standard over time for existing customers."
However, Rackspace said there would be "no forced migrations. Customers can transition to performance at their own pace."
The new servers will be available within the company's Dallas, Chicago, and London regions "later in November," the company said, and will arrive in Sydney and Hong Kong within the first half of 2014. ®
Sponsored: Hyper-scale data management