If you are looking for the ideal modular architecture for your datacenter compute / network / storage solution, there is a natural affinity for the increasingly popular blade servers and the Datrium DVX storage solution. Here’s why:
Blades: Not enough drives for Hyperconverged (HCI)
You probably shopped hyperconverged to some extent over the last year, and one thing you found was that drive bay counts on most blade servers (2) makes them unfit for HCI scenarios.
HCI configurations use at least 3 drives, and as many as 24 per server node. Picture this as seats in a vehicle.
Blades: 1-2 seats
HCI: 3-24 seats
In Datrium DVX storage configurations, 1-2 SSD drives in the ESXi host is a perfect – and recommended – solution for enhancing the compute & storage power of your hypervisor systems. It’s important to note that the Datrium solution is not limited to just 2 SSD drives but it’s more than enough on average. Recently, 2 SSDs on leading servers can mean more than 7 TB of raw flash, or as much as 40TB effective per server with the DVX.
This is OK because DVX host flash is just a massive cache. If a server or SSD fails, data is safe elsewhere.
Put SSD closer to CPU for lower latency
One of the key advantages of SSD is reduced latency, so putting it on the host would avoid network hops entirely for a majority of (read operations) storage IO in your applications.
In Datrium DVX configurations, you can populate the hosts with variable amounts of SSD (800GB – 8TB) sufficient to keep all data in cache and as close to the CPU as possible. The NetShelf provides a method for storing larger amounts of safe, shared and durable data across multiple hosts/blades.
Blades are powerful and DVX host overhead is too small to matter
The latest blade servers available today have significant horsepower and memory capacity and provide substantial and often a little extra compute in a very small footprint (e.g., CPU @ 32 core systems @ ~3 GHz and RAM @ 256GB to 512GB+ or about 8GB to 16GB per core).
Datrium DVX configurations, with the hypervisor CPU allocation at a modest 2 core (minimum) up to 20% and 10 core maximum of the available ESX host CPU, and less than 10GB RAM the Datrium DVX storage can obtain significant local storage performance into the thousands of low latency IOPS.
Buy SSD at server pricing, not servers at array pricing
When acquiring or updating blade servers, you will find that server flash is also a cost effective method for acquisition of SSD into your data center. Arrays and HCI embrace new drive technologies more slowly, and they charge array margins for you to use them.
Datrium DVX performance is based on the resources available at each hypervisor host but has no per-host or per-SSD charge. For the SSD configuration available in that host, drives are leveraged as a local cache pool using economies that can be realized – server by server – that are not possible with other traditional storage array or HCI approaches.
If you are looking to leverage the best features of modern data center technologies and blade servers are on your list, then take a close look at how the Datrium storage solution can leverage that investment even more.