SERVERS MOVE TO SINGLE SOCKET

It seems that the age of dual processors on server blades is giving way now to single CPU with more cores. The laptop here is able to handle the site traffic easily so far and its not likely a different machine is really needed.

See the source image
NVMe SSD server

Modern server processors have 8 memory channels to support 16 DIMMS which can afford a lot more RAM. This reduces the need for dual socket solutions of old. The latest models can support 4TB of RAM per socket. This can achieve 8TB with a dual CPU server and 16TB with a dual socket server.

Blade servers are still widely used and web servers can make do with less powerful servers. Database servers are best serviced with higher performance SSD arrays to allow the highest IOPS possible.

Even the IBM mainframe today is just another blade server albeit with a backwards compatible instruction set.

The trend towards one processor per blade is largely due to the chiplet designs. The new design makes for more performance on one socket than a dual socket box even 5 years ago. The design also uses less power and the new single socket blades are less costly. So new servers are changing the way of old with better performance for the power used.

Storage servers vary with perfmance substantially. The high capacity hard disk servers tend to be relativly slow driven by the design of them. Better grade servers with more controller logic are far faster,

See the source image
Lenovo Blade Server

The NVMe SSD server typically have 24 U.2 SSD devices across the front to allow for easy access. There is typically a crossbar to reduce the lanes required as most server possessors do not have enough PCI Express lanes. Typically these servers are connected to the blade servers with fast networks.

Old stype blade servers are now giving way to 1U higher perfmance servers. Many 1U servers can have 4 processors to handle demanding workloads. Remaining blade servers are now heavily discounted.

More recent servers are now using SSD formats similar to the M.2 seen in consumer systems but they are longer to afford more room for NAND chips. In practice the 2U server format is standard for storage platforms.

A lot of servers are using the underpowered Ceph to handle a storage array. The problem is that it does not handle a complete picture of the resources in a large data center. Lustre is built for supercomputing so it can better handle the server cluster.

%d bloggers like this: