Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

We are using MLC SSD's in production for our Cassandra clusters. It is very important to consider your workload when implementing SSD's in a production application.

There are MLC SSD's and SLC SSD's. The former being reasonably priced (think <1000$), and the latter costing in the neighborhood of 5-10k per drive.

MLC SSD's have a fixed write lifetime and will die or degrade after a certain number of writes. SLC SSD's have no such limits.

Cassandra for example has a very predictable write workload given our application, but other datastores (like an active write heavy mysql OLTP DB) would kill an MLC SSD very quickly.

We are using Cassandra to drive an API that must service thousands of requests a second but also with sub 100ms response times. Not a typical web use of Cassandra. Response times from hard disk just were not fast enough for our needs.

A new type of MLC is hitting the shelves now too. Enterprise MLC... supposed to be more resilient to the issues MLC ssd's have faced.

We are using OCZ Vertex.



I don't know about OCZ Vertex, but I remember that Intel bragged that its MLC drives can last for about 5 years even if you write a couple of GBs every day.


We are writing quite a lot more than 10GB per SSD per day on ours, and we have run into lifecycle issues. Of course, we are running a lot of these SSD's and striping across them with OS level raid, so we are doing a lot of write IO.


> SLC SSD's have no such limits.

[citation needed]




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: