Understanding radosgw benchmarks

We've noticed that there are a few common misconceptions around radosgw performance, and we're hoping that this post can clear up some of those.

BIT.group selects hastexo for Ceph training

"We're 2 hours into a 3-day class, and you've solved our most pressing production issue!"

BIT.Group is a professional services provider with offices in Germany and PR China, and a headcount of more than 300 employees. An SAP Gold Partner, BIT.Group has a long-standing cooperation with SAP AG to provide top-notch SAP solutions to customers in EMEA and across the globe.

BIT.Group operates BIT.Cloud, an Apache CloudStack based multi-tenant private/public cloud environment based on Ceph distributed storage.

In a three-day training engagement, hastexo Principal Consultant and instructor Florian Haas worked with a team of BIT.Group storage experts to expand the team's Ceph knowledge and optimize the system for performance and stability.

Spreading knowledge: OpenStack and Ceph in New Zealand and Germany

Lately, we've had the pleasure of working with two very interesting and capable customers, both of which were just on the verge of a product launch: Catalyst IT in New Zealand and SUSE in Germany.

In January, just after linux.conf.au 2015, we partnered up with Catalyst IT in New Zealand to deliver back-to back OpenStack training in the beautiful cities of Auckland and Wellington. A mixed crowd of Catalyst customers and employees went through our Cloud Fundamentals, Networking, Ceph and Security courses on the OpenStack platform. For us, it was an excellent experience to work with an extremely sharp Kiwi crowd that set aside some time in the New Zealand summer to learn about the world's fastest-innovating cloud platform. 

We taught our two-day Cloud Fundamentals course both weeks, adding two extra days with alternate modules each week. In week 1, attendees went straight into our OpenStack Neutron deep-dive, Networking for OpenStack (HX102). HX102 is a dense, fast-paced, highly technical 2-day deep dive into software-defined networking, OpenStack Neutron, OpenFlow and Open vSwitch, and also includes advanced Neutron topic like LBaaS, FWaaS and VPNaaS. The class was so capable and eager that we even had some extra time to cover some OpenStack Heat content from our Orchestration and Scaling (HX107) class.

In week 2, the add-on modules were our courses about Ceph Distributed Storage for OpenStack (HX104) and Advanced Security for OpenStack (HX108). This is where we provide a deep introduction to Ceph integration with OpenStack, and cover a bunch of critical OpenStack security topics.

A couple of weeks later, Catalyst IT officially launched the Catalyst Cloud, New Zealand's first public cloud service, based on OpenStack. Congrats, Catalyst! We're happy to have made our small contribution.

At the same time that Catalyst Cloud launched, we were already on-site with our next major customer: at SUSE in Nuremberg, we helped a team of support engineers and consultants hone their expertise on Ceph. Again, we were working with a sharp and demanding crowd, multinational this time: we had folks from Germany, France, Senegal, Spain, the U.S., the U.K., and the Czech Republic in attendance.

We ran an extended version of our Ceph Distributed Storage Fundamentals (HX112) class — normally a 2-day course, customized and extended to 3 days. (I should perhaps mention, by the way, that we teach this course on Ubuntu 14.04 LTS, SLES 12 and RHEL/CentOS 7.) Attendees got a deep technical introduction into the fundamentals of distributed storage, into RADOS, CRUSH, RBD, and RadosGW. We also covered Ceph integration in OpenStack (from our HX104 course) and even had time for a hands-on CephFS preview. Feedback from attendees was extremely positive with several reporting it was the best professional training class they had ever taken.

Shortly after, SUSE released SUSE Enterprise Storage, and again, we're thrilled about having made our own small contribution toward their product release. 

Congratulations again to both of these wonderful hastexo customers, and all the best with the new products — we'll see you again soon!

And just in case your company needs top-notch OpenStack or Ceph training: we're always happy to help. Just get in touch!



Ceph Performance Optimization (HX113)

The hastexo Ceph Performance Optimization (HX113) module is a 1-day course focusing on optimizing performance and stability for your Ceph distributed storage platform. It builds upon the Ceph Distributed Storage Fundamentals (HX112) course.


The class is highly interactive; any theoretical introduction to a new concept is accompanied by a hands-on lab.

Ceph Distributed Storage Fundamentals (HX112)

The hastexo Ceph Distributed Storage Fundamentals (HX112) module is our 2-day deep dive into the Ceph distributed storage platform. This course is designed to provide an in-depth understanding of Ceph's architecture, design, and implementation.


The class is highly interactive; any theoretical introduction to a new concept is accompanied by a hands-on lab.

Catch Martin Loschwitz and Syed Armani talking about Ceph this week!


p>Just coming off Percona Live 2014, we're speaking at two more conferences this week: the Open Source Data Center conference in Berlin, and the OpenStack India meetup in Mumbai. We'll have two of our best speakers covering Ceph at both of these events.

OpenStack & Ceph: A perfect match?


p>An overview of Ceph integration with OpenStack, presented at Ceph Day in Frankfurt, Germany (February 2014).

Fun with extended attributes in Ceph Dumpling

This is a rather nasty bug in Ceph OSD, affecting 0.67 "Dumpling" and earlier releases. It is fixed in versions later than 0.70, and a simple workaround is available, but when it hits, this issue can be pretty painful.

Please read this post to the end. This is by no means a punch being thrown at Ceph, in fact it rather clearly illustrates a very sane choice that the Ceph developers have made. If you run Ceph Emperor or later, you are not affected by this issue, but it will be an interesting read in data integrity in distributed systems anyway.

Unrecoverable unfound objects in Ceph 0.67 and earlier

As Ceph author Sage Weil points out frequently, distributed storage solutions for all their goodness have a "dirty little secret": No matter just how redundant and reliable they are by design, a bug in the storage software itself can be a real issue.

And occasionally, the bug doesn't have to be in the storage software itself.

OpenStack Italia User Group Meetup, Milan, June 20

Last week, I did a quick trip to Milan in order to speak at the fourth OpenStack User Group Italia meetup, hosted at Coworking Login by local ISP Enter.

Syndicate content