A new crop of systems are coming to the data center, and they owe a lot of their heritage to Linux. Space and power are at a premium in today's datacenter, and Linux will be the top choice for smaller and less power-hungry machines powering cloud computing.
A couple of systems have made the news recently that are targeting "scale out" cloud workloads, and Linux is likely to be the OS of choice on these systems. The SeaMicro SM10000 and the Quanta S2Q take different approaches, but the theory is the same: Cram a lot of low-powered cores into a system to handle workloads with massive amounts of small transactions. Sounds like a job for Linux!
The Quanta S2Q
According to Andy Patrizio on ServerWatch.com, the S2Q is a low-power and high-density machine that utilizes massively multicore CPUs. The S2Q is based on the TILEPro64, a RISC-based CPU with 64 cores on a chip running at 700MHz or 866MHz. According to ServerWatch, those cores are about as powerful as an Atom CPU.
The S2Q is rated for eight TILEPro64 chips, providing a whopping 512 cores per 2U server, addressing up to 64 DIMM slots. Targeted at cloud computing workloads, the S2Q won't run off-the-shelf software. It doesn't support the x86 instruction set, so software will need to be recompiled to run on the TILEPro64 chips. Bad news for Windows, good news for Linux. Not only is Linux tailor-made to work with cloud loads, it won't be hard to get a LAMP stack up and running on these systems.
The SeaMicro SM10000
What do you get when you combine 512 Intel Atom processors in one system? It's not the lead-in for a joke, it's a new approach to saving space and power in the data center. SeaMicro, a company focusing on reducing power consumption for data centers, has rolled out a new system that takes up less space, less power, and is tailor-made for "scale out" workloads.
The first system from SeaMicro is the SM10000. The SM10000 is a server with 512 Atom Z530 CPUs, 1 Terabyte of RAM, up to 64 SATA drives, up to 64 Gigabit Ethernet ports. All that fits in 10U of space, which means that you can fit four units in a standard rack. According to the company, the system reduces power consumption by 75 percent when compared to forty dual socket quad core 1U systems and support hardware.
Even taking the vendor numbers with the requisite grain of salt, the SM10000 does look like it offers some interesting advantages over traditional systems.
Use Enough Dynamite There, Butch?
The current crop of server-specific CPUs are way more horsepower than necessary for many use cases. If you're doing serious enterprise data transactions, you might need the latest 8-core CPUs. Companies that are handling heavy Web traffic, and thus a lot of small files, don't need the same kind of processing power just to churn out Web pages.
Even underutilized, those beefy systems suck up a lot of power. The Atom CPUs, while less performant, also consume far less power. Expect to see companies exploring similar solutions with ARM and MIPS CPUs soon. The Atom is likely to remain the leading choice, though, since its x86 architecture means that most organizations will be able to run their existing software unchanged.
Aside from the swarm of Atom CPUs, the actual design of the system is what makes it really interesting. The SM10000 combines the servers with the networking infrastructure. The system includes the rack switch, terminal server, and load balancer in one tiny case.
Why It's Important
The SM10000 is nice and all, but it clocks in at more than $130K. Pricing isn't announced for the S2Q yet, but it's not unreasonable to assume it won't be cheap. Is it really worth the sticker shock to save money on some space and power?
In a word, yes. Companies that run their own data center soon find that as their computing needs expand, so do the space requirements. Eventually, you're running out of room and looking at either expanding the data center (not cheap) or investing in smaller systems. If you're renting space in a data center, investing in a smaller system that fits in 10U of space could result in significant savings as well.
And the power requirements are a major factor. Google did a study in 2007 that indicates that the cost of powering systems could cost more than the actual equipment over the life of the systems.
Cost isn't everything. Some facilities are simply reaching capacity in terms of space and power availability. At that point, it doesn't matter if the power costs are excessive or not — you have to deal with the space and power available.
Not for Everything
These machines aren't going to be replacing commodity x86-64 servers for every use case. The servers are specifically designed for "scale out" workloads, and might not be appropriate for many types of computing. While the SM10000 may be ideal for cloud computing workloads, it may not be the ideal replacement for financial institutions or other workloads that require larger transactions. The TILEPro64 will require software to be recompiled, so it'll be ideal for LAMP applications — but not for proprietary software without source code available.
The SM10000, S2Q, and similar systems also have a few disadvantages. Commodity x86 and x86-64 systems may slurp down more power, but they have the benefit of being commodity systems. Even if your vendor goes out of business tomorrow (highly unlikely if you're getting systems from Dell, HP, etc.) you should be able to get replacement parts for those systems.
Since these systems are composed of non-standard parts, CIOs have to decide whether to take the risk on non-standard systems. If SeaMicro or Quanta don't make the long haul, what to do with those systems if a vital component goes belly up? You're not just losing a server, you might lose the entire system.
Approaching an Atom Age?
Are low-powered chips going to take over the data center? It seems unlikely, but I do expect to see variations on these systems carving out a niche for Web hosts and companies with the "scale out" workloads.
These systems will need to prove themselves in the datacenter before they'll become common-place. But the SM10000 and S2Q could be the fore-runners for a new breed of Linux-powered systems that will scale-out loads in data centers in the near future.