Red Hat Ceph Storage 3.2 brings a new round of enhancements

We are pleased to announce today’s immediate availability of Red Hat Ceph Storage 3.2, our fifteenth RHCS release. There are several highlights with this version, the most notable of which is the introduction of full support for BlueStore. As you definitely already know, BlueStore delivers very significant performance improvements to all Ceph users.

BlueStore

Red Hat Ceph Storage 3.2 introduces GA support for the next-generation BlueStore backend. BlueStore delivers a 2X performance improvement for clusters that are HDD-backed, as it removes the so-called double-write penalty that IO-limited storage devices (like hard disk drives) are most affected by. Additionally, BlueStore provides significant performance enhancements in all-SSD configurations, as detailed in our recent whitepaper with Micron. The key highlights of an...

Revisiting the Mustang board two years later, we learned that a few things have changed, and use three of them to build a test storage cluster to try Ceph on ARM — doing a little benchmarking while we are at it.

A New Breed

A newer revision of the Mustang EVK X-Gene 1 development board has relocated some of its ports and added a few more capabilities. The classic board we discussed at length in a previous blog is now identified with the model number 3PBx, whereas the new ones are part of the 3PCx series.

The SATA ports have moved, as you can see in the picture stacking a new board over the older one, and two ethernet ports and the SD card slot have traded places. Besides these cosmetic changes, an additional serial port has been brought out to a connector as UART1 and placed right above good old UART0.

A more substantial change is the adoption of a new UEFI firmware made by...

Kyle Bader and I teamed up to deliver a quick (and hopefully painless) review of what types of storage your Big Data strategy needs to succeed alongside the better-understood (and more traditional) existing approaches to structured data.

Data platform engineers need to receive support from both the Compute and the Storage infrastructure teams to deliver. We look at how the public cloud, and Amazon AWS in particular, tackle these challenges and what are the equivalent technology strategies in OpenStack and Ceph.

Tradeoffs between IO latency, availability of storage space, cost and IO performance lead to storage options fragmenting into three broad solution areas: network-backed persistent block, application-focused object storage (also network based), and directly-attached low-latency NVME storage for highest-performance scratch and overflow space.

Ideally, the infrastructure...

Los Tres Caballeros —sans sombreros— descended on Vancouver this week to participate in the “Rocky” OpenStack Summit. For the assembled crowd of clouderati, Sébastien Han, Sean Cohen and yours truly had one simple question: what if your datacenter was wiped out in its entirety, but your users hardly even noticed?

We have touched on the disaster recovery theme before, but this time we decided to discuss backup as well as HA, which made for a slightly longer talk than we had planned—we hope you enjoyed our “choose your disaster” tour, we definitely enjoyed leading it.

Our session on Ceph disaster recovery in OpenStack also featured unique parenting advice!

The recording of our Summit session is now live on the OpenStack Foundation’s YouTube channel. It is impressive how quickly the Foundation’s media team releases now:

Our slides are available as a PDF and can be viewed inline...

Neil, Uday and I joined forces to deliver a roadmap update at the annual Red Hat Summit, hosted this year at Moscone West in San Francisco. But before that, we started the week by running a meeting of the somewhat secretive (and glamorously mysterious) Ceph Advanced User Group, spending Monday hearing what our favorite think-tank had to say about our roadmap plans for the next year.

As always, our slides are available as a PDF download and can be viewed inline below.

Comments? Questions? Ask us on Twitter.

Cryptography made easy…er

Cryptography does not have to be mysterious — as author of Serious Cryptography Jean-Philippe Aumasson points out. It is meant to be fiendishly complex to break, and it remains very challenging to implement (see jokes on rolling your own crypto found all over the Net), but it is well within the grasp of most programmers to understand.

While many are intimidated by the prospect of digging into what is effectively a branch of number theory, the reality is that cryptography is squarely based in discrete mathematics—and good coders are all, without exception and often unknowingly, natural discrete math jugglers. If you are interested and you aced your data structures course, chances are that crypto will not be an unsurmountable challenge to you. Aumasson certainly seems to think so, and he walks us along his own path to the discovery of the cryptographic...

The third one is a charm.

RHCS 3 is our annual major release of Red Hat Ceph Storage and it brings great new features to customers in the areas of containers, usability and raw technology horsepower. It includes support for CephFS, giving us a complete all-in-one storage solution in Ceph spanning block, object and file alike. It introduces iSCSI support to provide storage to platforms like VMWare ESX and Windows Server that currently lack native Ceph drivers. And we are also introducing support for client-side caching with dm-cache.

On the usability front we are introducing new automation to manage the cluster with less user intervention (dynamic bucket sharding), a troubleshooting tool to analyze and flag invalid cluster configurations (Ceph Medic), and a new and rather impressive monitoring dashboard (Ceph Metrics) that brings unparalleled insight into the state of the cluster.

Wherein the three musketeers travel to the antipode, find a fourth musketeer, and deliver a new talk in spite of jet lag

Sébastien Han and yours truly traveled to Sydney to meet with the gathering OpenStack Community and our very own D'Artagnan: Monsieur Andrew Hatfield.

Discussing the progressive integration of OpenStack with containers, our talk spanned Linux, Docker, Kubernetes, and Ceph for maximum headache. Ansible and Triple-O tooling rounded up the whirling technology carousel — at least for seven minutes, until we introduced Kolla to a dizzy audience!

Thanks to the time-zone advantage, I also had the chance of being the first one globally to announce Red Hat Ceph Storage 3.0, launching today — all hosts are storage hosts now, with containerization!

UPDATE The media team has now posted the live recording of our session in the OpenStack Foundation’s YouTube channel.

...

The point of UX is making things effortless

I usually resist sharing at large my opinions on how to make software, out of some impostor syndrome nonsense modesty thing. But this was too much fun not to write it up and share with you all.

User experience is key to the success of any software product today. Yet I find that too many folks in our industry do not really understand what UX truly is. Let’s set aside software for a minute, and illustrate this with an example from a lofty place.

Trash Talk

This trash bin was clearly designed by a well-meaning engineer. It accounts for every possibility… and it makes you play 10 questions every time you approach it. Are the forks compostable? Yes, turns out they are in this building. As in all cases where UX is involved, the point is not that saving the environment (or getting your network settings right) is not worth the effort. The point...

Hardware hacking 101 returns to MIT’s BLU

After a year of inadvertent absence from my favorite Linux user group (turns out that taking over the world with Software-Defined Storage is time consuming!), I returned to BLU to review my newest Linux cluster designs.

I reviewed design options for four clusters using x86 and 64-bit ARM processors housed in systems with a range of prices but with an eye to accomplish each objective with the minimum dollar cost: OpenStack cluster, Docker cluster, Ceph storage cluster and a Raspberry PI cluster — because of course everyone must have a Raspberry PI cluster! More seriously, it is actually a really useful lower-bound for minimal cluster cost.

We looked at the design considerations that went into each design, spanning from four nodes to twelve, and from $150 to $3,495-ish, and inspected on stage the hardware of the two clusters I managed...