Pack the cluster in a box!

    Recently, the materials were all about the network and about the network, it is necessary to dilute the trend.

    Two servers and a JBOD disk shelf are a minimal set that allows you to build a fault-tolerant solution. At a minimum, these are 4U rack spaces, additional power cables, and server connections to storage. All this adds complexity to the initial installation and configuration of the system.

    To simplify the solution, reduce power consumption, reduce cost and required rack space, a new class of systems has been developed - a cluster in a box. A single chassis includes two servers, fault-tolerant power, cooling, and single disk space subsystems.



    With this approach, the solution does not require special conditions in the server room, special knowledge in the field of fault tolerance and clustering, since all connections have already been laid. All that needs to be done during installation is to connect the system to power and the local network.

    Examples of this solution are our ETegro Hyperion RS420 G4 and Hyperion RS430 G4 products.



    The Hyperion RS420 G4 model is made in 2U format for two servers with a single storage:



    On the front side there are 12 hot-swappable drives in universal disk baskets that allow you to install SAS disks of sizes 3.5 "and 2.5". The servers are turned on independently, and a button with the System ID LED is also displayed.



    Each node supports two Intel Xeon E5-2600 v2 processors (up to 130 W, allows you to use any model in the series), 512 GB of RAM (16 slots), two 10G Base-T Ethernet ports, two internal seats for SATA drives for installing the OS It has a dedicated service processor and a dedicated Ethernet port for remote management.

    For communication between nodes, there is an internal Gigabit Ethernet channel routed through the internal backplane. This implementation allows you to organize the synchronization of the cluster ("heartbeat") without the use of external interfaces.

    The system has free PCI-E slots for installing additional expansion cards, such as additional network cards or accelerators based on PCI-E flash, one slot is always occupied by the SAS controller.

    Cluster disk space is scaled both by an external SAS port on each node, and by installing an additional SAS controller to connect new cascades of disk shelves.

    Compared to a cluster of 2 separate servers and an external disk shelf, energy savings reach 40%, space savings of 50%, and cost savings of up to 20% when configured with 12 drives.

    Hyperion RS430 G4 Server Update A general overview of the Hyperion RS430 G4 server

    was once published . Two dual-processor servers with 10G SFP + in one box, 35 disks each - a product specific and popular strictly in certain circles.

    I would like to expand the popularity, so we thought and came up with a natural change (in addition to production improvements), which we will talk about now.



    The most obvious is the replacement of plastic handles for extracting backplanes with metal levers, which greatly simplified the life of support and eliminated the need to keep a supply of plastic.

    The handles for removing from the rack were replaced by folding levers, which also slightly corrected karma.



    But these are cosmetic changes, the most important change is inside.

    What can be redone in a system where all components support hot swapping, and disks are divided equally between servers 35-35?
    We wrote a lot about the concept of Shared DAS and the delights of moving away from traditional storage to direct disk management and a logical change would be to stop sharing disks between servers and make them all available to each server as 70-70.

    Fortunately, the structure allows us to do this with "little blood" - all communications are carried out through printed circuit boards, and it is enough to change only one component so that access to each expander is failed from each server.



    In connection with the use of four expanders, at least 16 SAS ports are required in each server - these are either two sets of SAS HBA, or two sets of LSI Syncro .

    Now the server is suitable not only for storing everything and everything, but also for building fault-tolerant solutions on:


    In our line of Shared DAS laboratory solutions arrived :)


    Also popular now: