We host our sites with YottaServe, the first UK hosting operation to develop hosting solution products that deliver affordable ultra low server energy consumption - and supply it as standard. The technology behind the system was researched heavily from 2006 to 2007, and first saw live production in Q3 2007 - a new methodology for affordable, ultra low energy hosting was born.

Whilst creating the system YottaServe engineers identified a number of factors that contribute to energy waste in more typical hosting data centers. These key factors were identified as:

• Data Provisioning
• Server Chassis Hardware & Silicon
• Application efficiency

In their own words :

“Data Provisioning - Hard Disks

In most cases, all servers need hard disks, but when looking at typical server storage architecture, we felt that standard practice was very wasteful. Couple this with the fact that today, both SATA and SAS hard disks are spinning faster (7,500 rpms and 15,000 rpms spindle speeds are common), servers laden with hard disks were going to pull more current from the hosting rack, and of course generate more heat. The hotter computer components get, the less efficient they become, so the harder they have to work. Excess heat also stresses components and makes them more likely to become error prone and fail.

A standard specification for a server which is serving web pages, and a database would come supplied with two disks mirrored up as boot disks (RAID1) and at lest two more for client data, again, mirrored up in RAID 1. Mirroring hard disks is very inefficient, because in this above mentioned setup, the server is powering 4 hard disks, but only the storage from 2 is being used.

The answer was to remove the hard disks from the servers all together. By booting the servers off clustered Storage Area Network solution (SAN), the SANs clustered storage capacity can be ʻsliced offʼ and allocated to diskless servers and ʻgrownʼ as and when they need more space. This means that each diskless server has no mirrors or hard disk wastage. You therefore need far fewer hard disks, and because of this draw less power, and generate less heat.

The other benefit of removing the hard disk from the server is that if the server hardware were to fail, an offline spare could be powered up remotely, and could ʻattachʼ to the data volume previously used by the dead server. In the past, the only way to quickly (within minutes) recover from a ʻbare metalʼ server failure was to deploy servers in groups of two or more. The diskless method allows for one powered off ʻroaming spareʼ to be called up at any time, meaning fast recovery from catastrophic server failure, without having to double up on all your hardware. This not only saves power, but also huge amounts of cost in the addition space, hardware, software and finally management complexity of doubling up your hosting operation just on the off chance you might suffer hardware failure.

Server Hardware

Removing hard disks from servers enabled us to look at drastically shrinking the size of our servers down, as their only moving parts would now be fans, and without hard disks (which generate significant amounts of heat) would now run far cooler. These smaller servers not only take up half the space of typical ʻ1Uʼ servers, also use highly energy efficient PSUs, rating from as low as 230W. The use of ultra low power, dual and quad core CPUs are also important factors in keeping the resource footprint of the server as low as possible. The Intel XEON 5410 quad core 2.3Ghz CPU draws just 50W, and cost more then their more power hungry variants. Server nodes are also configured to enter special low energy modes when in periods of low workload - at night time for instance.

Application optimization & acceleration

The third component of YotEnv, this plays a significant role in drastically reducing the servers processor % utilization. It is no good having ultra efficient servers, if they are running applications which cause the servers CPU to run at 100%. Even if the CPU is of the ultra low voltage variety like our XEON 5410s - 100% CPU is going to require more power, and generate more heat, than the same server running at
5% CPU utilization. Lab tests show that with the application acceleration technologies we use, CPU % utilization can be dramatically reduced.

Server Virtualization

Once the underlying hosting methodology has been created, it is then possible to apply other further energy saving technologies such as ʻ virtualizationʼ - where multiple virtual servers can be run on one hardware node. Unless virtualization systems are enterprise class (clustered), the drawback of placing multiple virtual servers on a single hardware platform, heighten the risk of data loss from a server failure if hardware running 5 virtual servers were to go down, then you would lose 5 servers not one.

YottaServes virtualization platform is of enterprise specification and, thanks to YotEnv, runs even more
efficiently that traditional virtualization setups. Loss of a server node in our setup instigate an automatic procedure whereby any failed virtual servers would start up on nodes with spare resources.

We are also further developing new energy saving methods supported by the virtualization platform. For instance, at night time, virtual servers may be automatically moved to fewer nodes, and any remaining nodes be powered off or set into ʻP stateʼ - all with zero downtime. When servers start getting busy in to morning, nodes are powered back on, as and when the resources are needed.

Our aim is to provide reliable, responsible, and highly responsive hosting solutions, and it is our YotEnv methodology that enables us to achieve this.

The power draw on YottaServes hosting centre is currently around the 4 - 5 AMP mark, and runs 9 servers - on the previously run setup, the same configuration would have needed in excess of 10 AMPS, double the amount of energy required to power the hardware.”