InMotion Hosting fixes shared hosting crashes, replaces Linux component most providers never touch

Shared hosting instability has a familiar pattern that most customers know well. The WordPress dashboard slows down, the database throws an error, and the site goes offline briefly before recovering. Support then tells you MySQL consumed too much memory, or a plugin misbehaved, or traffic spiked at the wrong moment. The server reboots, things stabilize, and everyone moves on without a clear answer.

InMotion Hosting, however, spent time actually looking for one. What its engineering team found was not in the database configuration or the application layer. Instead, it sat deeper, in the default Linux memory allocator that ships with most Linux distributions and handles how memory gets requested and released at the system level.

The glibc malloc allocator, which most hosting providers run without modification, gradually fragments memory under sustained load. On a shared hosting server running hundreds of websites through a single database layer, that fragmentation builds over weeks. As a result, memory consumption climbs steadily until the system hits its ceiling, Linux starts killing processes to protect stability, and customers see crashes that look random but follow a predictable underlying pattern.

To address this, InMotion replaced glibc malloc with TCMalloc, a memory allocator originally developed by Google for large-scale production environments. The swap happened fleet-wide across the company’s shared hosting infrastructure. Furthermore, over a two-week monitoring period following the deployment, the gradual memory growth pattern that previously triggered forced resets reportedly disappeared entirely.

The results are significant. MariaDB memory usage dropped more than 50 percent across the fleet, with some heavily loaded servers showing reductions closer to 77 percent. Sean Combs, Tier 3 Systems Administration Technical Lead at InMotion, noted that gradual MySQL memory growth often looks identical to a database memory leak until engineers examine allocator-level behavior directly. Consequently, the underlying cause went unaddressed for so long because the symptoms pointed elsewhere.

Erik Soroka, Director of IT and Data Center Operations, framed the project as the kind of infrastructure work that customers experience through improved stability rather than anything visible in a control panel. That framing is accurate. Most shared hosting customers will never know what a memory allocator is, or even that their provider changed one.

Nevertheless, what they will notice is fewer crashes, more consistent database performance, and support tickets that do not end with a shrug and a reboot.