Our Compute cloud offerings run completely on Open Source software. As a cloud orchestrator we make use of Apache CloudStack and for the redundant storage behind our High Availability offering we use Ceph, an Open Source scale-out distributed storage platform. Ceph allows us to create highly available and reliable storage behind our High Availability cloud product, which we can expand without downtime.
We have been using Ceph in production since 2013 and have been loving it ever since. Recently, we have started upgrading our first clusters to the latest Ceph release (Luminous) and we wanted to share what this new release brings for us, but also for you, as our end user.
Stable and Major releases of Ceph all get a unique designation. It started with Argonaut, which was superseded by Bobtail, Cuttlefish, Dumpling, Emperor, Firefly, Hammer, Infernalis, Jewel, Kraken and now Luminous.
At PCextreme we started using Ceph at the time of the Dumpling release and have been upgrading ever since, implementing Firefly, Hammer, Jewel and now Luminous.
Initially Ceph used the XFS filesystem to store data, utilizing a storage backend called FileStore.
In essence this worked fine but it had various drawbacks when it came to performance and reliability. This has lead to the development of a new backend called BlueStore,
which consumes a RAW block device to store data on instead of using XFS.
While a boost in performance is always nice, CRC on checksums on all data ensures even more reliability. Fortunately, during our 5 year span of running Ceph, no data has ever been lost or corrupted. That being said, having additional safeguards is always wise.
So, in a nutshell, what makes BlueStore better then FileStore?
- 40% performance improvement
- Active CRC checksums on all stored data
In addition to BlueStore, the Luminous Release brings a lot more features to the table.
Although our Ceph clusters operate using reliable hardware, they do fail every so often. Hard drives, SSDs and Memory are all vulnerable to failure.
Luminous detects and handles these failures even smoother than the previous release did (Jewel). As a result, if a storage component powering your High Availabilty server happens to fail, you will barely notice it.
The Luminous release also uses less resources such a memory and CPU. This means snappier I/O and less latency, something we all enjoy very much.
Hopefully, you haven’t even noticed when we upgraded Ceph to the Luminous release. We've been able to do this seamlessly in the background without downtime!
Many hard disks/SSDs powering the platform also have been converted from FileStore to BlueStore without interruptions.
Needless to say we are very happy with the Luminous release and we're looking forward to what the future will bring next time.