Spectre Meltdown

Spectre and Meltdown will be the Death of On-Prem Servers

Recently revealed bugs in CPUs have caused a 25–35% slowdown in processing speeds for server farms. To understand this, we need to understand a bit of processor architecture.

Processors have the ability to predict the outcome of a transaction, and prepare themselves for it, known as branch prediction. Imagine you’re driving home, you’re pretty sure your house is in one direction and you start driving before you look at a map. You making that guess is similar to how the processor creates a branch of a transaction. If you, or the processor, is wrong you would have to back track to where you were, then make the right decision. Usually the processor is correct, allowing for huge savings in time.

Spectre and Meltdown take advantage of this branch structure. Meltdown creates fake branches that “meltdown” the barriers in between branches and allow data to be read that should be independent. Spectre peers from one branch into another, similar to a creepy ghost inside a machine.

These flaws are in AMD, Intel and Qualcomm chips, spanning the vast majority of servers, desktops and mobile devices. Meltdown has been mostly contained, however it seems Spectre takes advantage of gaps between the virtual machines and the hypervisor in servers and can use Javascript to read unauthorized information. We are still in the early days of this bug, and although it is hard to exploit, it is attainable.

A huge benefit to using public cloud providers such as Google, Amazon and Microsoft is they are constantly monitoring and researching these vagrant security threats. We are in the early stages of this bug so now is a great time to consider moving legacy systems to the cloud, and take advantage of cloud provider security resources. On-premises hardware would also be rendered obsolete almost immediately if Spectre and Meltdown cannot be overcome by software patches.

At Caserta, we have seen a slight increase in our Spark cluster size since December 20th, however throwing a few more machines at the problem is an easy fix. We also advise containing these clusters inside a VPN to ensure data protection. We will be monitoring this threat carefully and look forward to working with our partners to protect their data.