Category Archives: Software

Enterprise server as a desktop?

January 3, 2017 ioko Leave a comment

We may all have seen used servers sold in the ebay for few hundreds of bucks. Highly equipped with Xeon processors, many GBytes of RAM, professional RAID controllers and ultra fast SAS disks. The question is if such a server can be used as a desktop – home PC for our regular needs.

Some of the factors that we should examine before deciding on with hardware we should invest are the followings:

Noise, acoustic noise may be the most critical part in selecting the appropriate mode. Servers have designed to work under controlled environments without humans nearby. In addition, low profile cases, in order to provide the required air flow, usually use fans running in extremely high RPMs and creating loud noise. You probably do not want to work close to box generating more than 32 to 35dbA. So, examining the specification of the server is essential. Alternative, special boxes can be used, equipped with noise absorbing materials, but always considering providing enough air flow.

Power consumption, the rule is “the newer, the best”. Every new CPU core that is released has better power consumption characteristics and more sophisticated power management features. Ie an old HP Proliant DL380 G5 consuming ~250Watts in idle for the 2 quad cores Intel X5450 CPU while a newer HP Proliant DL360 G7 (~6 years old) consumes ~90Watts for the 2 hexacore L5640 CPUs. A good estimation can be obtained by the CPU specifications concerning the TPD or by the spreadsheets provided by the server vendor – providing a detailed power and thermal profiling.

CPU performance, that’s what’s all about. Enterprise servers are equipped with multiple Xeon processors having huge L3 cache memories, multiple cores – usually four or six cores per CPU and in some cases more than 20 (!) – and hundreds on GBytes RAM memory to support the tens of cores. For sure, used server – sold for few hundreds bucks – are not equipped with the latest – state of the art – CPUs, but let’s compare a 5-year old enterprise CPU with a high performance latest desktop CPU. Intel X5650 CPU is commonly used in enterprise servers and you can have in a reasonable price – with some servers sold as low as 300 to 400 $. In this price range we may have a new i7-6700 CPU:

*CPU*	*Dual X5650*	*I7-6700*
*Frequency*	2.66/3.06GHz	3.4/4.0GHz
*Cores (Phy/Vir)*	2 * 6/12	4/8
*L3 Cache*	2 * 12MB	8MB
*TDP*	95W	65W
*PassMark Score*	11708	10024
*Price*	Obsolete	~400$

Size, the dimensions of a rackmount server is not always what we have used in desktop PC. We should make sure than the size needed on our desk is available for our new monster. An alternative way is to drill the desk and mount the server vertically, as you can see in the following photo, it this way the noise will be moreover reduced, the server will not be visible, no extra space required and no extra cost required:
GPU, although the servers are already equipped with a VGA output, this can only be used for maintenance purposes and definitely not for running a desktop environment that we are going to work with. Luckily, various PCIx slots are available, so you can plug your GPU card and be able, not only to work, but also to run your favorite games. At this point, we should consider the available space for the PCI card inside the server, most 1U size server accepting only low profile PCI cards, the needs for external power, not all servers providing additional 12V power outlet to feed the GPU, and finally the ventilation of the card. A good choice may be the GT 730 nVidia card, that I also used in my system. A low profile card, with a good overall performance and without external power feeding needs. Additionally, you can choose a fan-less version, or you can make your own version as it shown in the following image:

GT 730 mounted on the PCI riser

Disks, many of the available servers are already equipped with used Serial Attached SCSI (SAS) disks. SAS disks are usually faster than the SATA – desktop disk – and more reliable. The high spinning RPM but also the high data transfer rate even for SAS-2 disk, making SAS disk an excellent option from performance perspective. In addition taking advantage of RAID technologies such as RAID 5 or RAID 6 that are already available on almost all the enterprise servers, we can further improve the performance – having better performance even from a single SSD disk (which may cost more than whole server). Also, this we reduce the risk for data loss due to disk failures (since disks are already used and maybe are already close to MTBF). Please note that in case that we are planning to use advanced RAID configurations, ie RAID 5 or RAID 6, we should check that the server is equipped with all the necessary modules. In some cases, additional cache and battery modules are required for configuring disk stripping or an extra license it might be required for enabling some options.

In conclusion, selecting the right option we may have a complete high performance used system in the price of a latest new processor (similar to performance CPU). In any case, we have to carefully consider the particularities of an enterprise server, be prepared for bidding the right item and get ready for launch!

Blog, Software

2017: the Big Data Year?

December 29, 2016 ioko Leave a comment

It is the first time that the storage space is almost unlimited. It is not only possible to have petabytes or even zettabytes of cloud storage, but also we can have it, in an affordable price. Cloud storage providers – such as Amazon – have reserved fleets of trucks, each one transferring hundreds of petabytes, or even better “tonnes” of data, every day to the cloud storage. Different kinds of information are stored and are accessible anytime from anywhere. IoT sensors information is transferred in real time from all over the world, business transactions are stored and are available for years, mobile devices are constantly uploading data of any kind that we can imagine: sensor data, images, videos, positioning information. Health devices are providing information from monitoring devices (ie pulse rate, BPM, SpO), information about the medication progress or symptoms of the patient but also information about the patients’ habits (ie daily steps or sleep quality). This information can be securely stored and provide a valuable tool for the researchers.

In addition, processing power cost has been dramatically reduced. Clusters with hundreds of CPU can be bought for few thousands of euros in ebay. Moreover, the processing power is located in the same place where the data is: in the cloud. Virtual environments, dockers, containers are available to accommodate any platform that is needed for serving our needs, optimising the data usage. Elastic allocation can provide us the resources needed the time needed. All the above are tied together in user-friendly platforms such as Apache Hadoop making possible to deploy and provision a new cloud platform only in few hours.

What we can expect from the big data for the new year?

“Big data” was rapidly grown the last few years and we are now in place to form the brain of the new internet era. Software development is transformed to software science. We do not care about the data as data, but instead, we now searching for the actual information included in the abundance of data, depending on our task. Advanced signal processing algorithms are used to extract information based on the requirement, correlate different types of information and furthermore being able to make decisions. Machine learning, neural networks or deep learning, as it used to be called in “big data”, making the software scientists the most highly paid professionals, constructing the internet brain. People living in big cities are already familiar with applications, like the traffic information provided by google maps. This is an excellent example of Big Data. We do not care about who when and why is moving, we just want to know the status of the traffic in a specific place. Information is available – from the thousands of devices moving in this area – and big data are responsible for processing the information giving us the status of the traffic in real-time. Moreover, we may predict the traffic for a specific time based on the historical data of similar days, period or even weather conditions. We may see internet taking part actively in criminal prevention, identifying risks and managing portfolios without any human interactions. Big data will become a part of our daily life as the 21st century Pythia.

Software