Gluster blog stories provide high-level spotlights on our users all over the world
Long time, no see huh!!! This post has been pending on my part for a while now, partly because I was busy and partly because I am that lazy. But it’s a fairly important post as it talks about snapshotting the GlusterFS volumes. So what are these snapsh…
Important happenings for Gluster this month: We’re closing in on a 3.8 release, with release candidate 2 released on May 24th. (http://www.gluster.org/pipermail/gluster-devel/2016-May/049642.html) Our 3.8 roadmap of features is available at: https://www.gluster.org/community/roadmap/3.8/ Our current timeline is to have a release in June, so update your release notes! Gluster Developers Summit: October 6, 7 directly following LinuxCon …Read more
Lots of people at Vault this week! Last month’s newsletter included a list of all of the Gluster related talks, so if you’re here, come by the Red Hat booth and say hi! New things: 3.7.11 Released: https://www.gluster.org/pipermail/gluster-devel/2016-April/049155.html We’ve got a new Events (https://www.gluster.org/events/) page to replace the publicpad. (https://public.pad.fsfe.org/p/gluster-events) – Feel free to add …Read more
This is first post in a series of blog posts where we will discuss deduplication in a distributed data system. Here we will discuss different types of deduplication and various approaches taken in a distributed data systems to get optimum performance and storage efficiency. But first basics of deduplication! “Data deduplication (often called “intelligent compression” […]
OpenShift is a platform as a service product from Red Hat. The software that runs the service is open-sourced under the name OpenShift Origin, and is available on GitHub. OpenShift v3 is a layered system designed to expose underlying Docker and Kubernetes concepts as accurately as possible, with a focus…
Great things happening this month! 3.7.9 release on March 22 3.8 roadmap and release planning Niels de Vos is kicking off a great deal of our work on 3.8 Previously, Automated Tiering in Gluster, Gluster at FAST, Gluster at NOS Upcoming for next Month: Linux Foundation Vault GlusterFS and its Distribution Model – Sakshi Bansal GlusterFS …Read more
This post describes how to run automated tiering in Gluster. Tiering is appropriate for stable workloads where frequently used data fits on small, fast storage, such as SSDs, and rarely used data resides on a slower/cheaper volume, such as spinning disks. On a tiered volume, files are tracked according to frequency of access. Popular files tend to migrate to faster storage, and unpopular ones to slower storage. The …Read more
We hosted a small meetup/birds of a feather session at USENIX’s FAST conference. FAST is a conference that focuses on File And Storage Technologies in Santa Clara, California. Vijay Bellur, Gluster Project Lead did a short talk on Gluster.Next, our ongoing architectural evolution in Gluster to improve scaling and enable new use cases like like storage as a …Read more
The last week of January and the first week of February were packed with events and meetings.
This blog contains my observations, opinions, and ideas in the hope that they will be useful or at least interesting for some.
The day before FOSDEM starts, the CentOS project organizes a community
meetup in the form of their Dojos at an IBM office in Brussels. Because
Gluster is participating in the CentOS Storage SIG (special interest group), I was asked to present something. My talk had a good
participation, asking about different aspects of the goals that the
Storage SIG has.
Many people are interested in the Storage SIG, mainly other SIGs that
would like to consume the packages getting produced. There is
also increasing interest from upcoming architectures to get Gluster
running on their new hardware (Aarch64 and ppc64le). The CentOS team is
working on getting the hardware in the build infrastructure and testing
environment, the Gluster packages will be one of the first SIG projects
going to use that.
I was surprised to see two engineers from Nutanix attend
the talk. They were very attentive when others asked about VM workloads
and hyper-convergence-related topics.
The CentOS team maintains a Gluster environment for virtual machines. It
is possible for CentOS projects to request a VM, and this VM will be
located on their OpenNebula “cloud” backed by Gluster. This is a small
environment with four servers, connected over Infiniband. Gluster is setup
to use IPoIB, not using its native RDMA support. Currently, this is
running glusterfs-3.6 with a two-way replica, OpenNebula runs the VMs over
QEMU+libgfapi. In the future, this will most likely be replaced by a similar setup
based on oVirt.
At FOSDEM, we had a very minimal booth/table. The 500 stickers that
Amye Scavarda brought and a bunch of ball pens imported by Humble and Kaushal
were handed out around noon on the second day. Lots of people were aware of
Gluster and many were not. We definitely need a better presence next
year, visitors should easily see that Gluster is about storage and not
only the good-looking ant. Kaushal and Humble wrote detailed
blog posts about FOSDEM already.
Some users that knew about Gluster also had some questions about Ceph. I
unfortunately could not point them to a booth where experts were hanging
around. It would really be nice to have some Ceph people manning a
(maybe even shared) table. Interested users should get good advice on
picking the best storage solution for their needs, and of course we
would like then to try Gluster or Ceph in the first place. A good
suggestion for users is important to prevent disappointment and possibly
The talk I gave at FOSDEM attracted (a guestimated) 400-500 people.
The auditorium was huge, but “only” filled somewhere between 25-50% with
a lot of people arriving late, and some leaving a few minutes early.
After the talk, there were a lot of questions and we asked to move
the group of people to a little more remote location so that the next
presentation could start without the background noise. Kaleb helped in
answering some of the visitors questions, and we directed a few to the
guys at the Gluster booth as well. The talk seemed to have gone well, and I
got a request to present something at the next NLUUG conference.
This was mainly informal chats about different topics listed in this Google Doc. We encouraged each topic to add a link to an etherpad where notes
are kept. The presenters of the sessions are expected to send a summary
based on the notes to the (community) mailing lists, which I won’t cover here.
Some notes that I made during conversations that were not really
Richacl needed for multiprotocol support, Rajesh will post his
work-in-progress patches to Gerrit so that others can continue with
his start and get it in for glusterfs-3.8. (Michael Adam)
QE will push downstream helper libraries for testing with distaf to
the upstream distaf framework repo or upstream tests repo. MS and
Jonathan are the main contacts for defining and enforcing an
“upstream first” process. “Secret sauce” tests will not become part
of upstream (like some performance things), but all basic
functionality should. At the moment we only catch basic functionality
problems downstream, when we test upstream we should find them
earlier and have more time to fix them, less chance in slipping
Downstream QA will ultimately consist out of running the upstream
distaf framework, upstream tests repo and downstream tests repo.
Paolo Bonzini (KVM maintainer) and Kevin Wolf (QEMU/block maintainer)
are interested in improved Gluster support in QEMU. Not only
SEEK_DATA/SEEK_HOLE would be nice, but also something that makes it
possible to detect “allocated but zero-filled.” lseek() can not
detect this yet, it might be a suitable topic for discussion during
LSF/MM in April.
One of the things that the libvirt team (requested by oVirt/RHEV)
asked about was support for “backup-volfile-server” support. This was
a question from Doron Fediuck at FOSDEM as well. It was the first time
I heard about it. Adding this seemed simple, and a train ride
from Brussels to Amsterdam was enough to get something working. I was
informed that someone already attempted this approach earlier… This
work was not shared with other Gluster developers, so the progress on
it was also not clear :-/ After searching for proposed patches, we
found that Prasanna did quite some work (patch v13) for this. He was
expected to arrive after the meetup with the virtualization team was planned.
Kevin did send me a detailed follow-up (in Dutch!) after he reviewed
the current status of QEMU/gluster. There are five suggestions on his
list, I will follow-up on that later (plus Prasanna and gluster-devel@).
Snapshots of VM images can be done already, but they would benefit
from reflink support. This most likely will require a REFLINK FOP in
Gluster, and the appropriate extensions to FUSE and libgfapi.
Something we might want to think about for after Gluster 4.0.
Finally, I met Csaba Henk in real life. He will be picking up adding
support for Kerberos in the multitude of Gluster protocols. More
on that will follow at some point.
Unfortunately, there was no Gluster swag or stickers at DevConf.cz, but this time there
were Ceph items! It feels like the Ceph and Gluster community managers
should work a little closer together so that we’re evenly recognized at
events. The impressions that I have heard, was like “Gluster is a
project for community users, Ceph is what Red Hat promotes for storage
solutions.” I’m pretty sure that it is not the message we want to relay
to others. The talks on Ceph and Gluster at the event(s) were more
equally distributed, so maybe visitors did not notice it like I did.
During the Gluster Workshop (and most of the conference), there was
very bad Internet connectivity. This made it very difficult for the
participants to download the Gluster packages for their distribution. So
instead of a very “do-it-yourself” workshop, it became more of a
presentation and demonstration. From the few people that had taken the
courage to open their laptops, only a handful of attendees managed to
create a Gluster volume and try it out. The attendees of the workshop
were quite knowledgeable, and did not hesitate to ask good questions.
After the workshop, there were more detailed questions from users and
developers. Some about split-brain resolution and prevention, others
about (again) the “backup-volfile-server” ‘mount’ option for QEMU. We
definitely need to promote features like “policy based split-brain
resolution,” “arbiter volumes,” and “sharded volumes” much more. Many
users store VM images on Gluster and anything that helps improving the
performance and stability gets a lot of interest.
Nir Soffer (working on oVirt/RHEV) wanted to discuss some more about
improving their support for Gluster. They currently use FUSE mounts and
should move to QEMU+libgfapi to improve performance and work around
difficulties with their usage of FUSE filesystems. At least two things
could use assistance from the Gluster team:
Speaking to Heinz Mauelshagen (LVM/dm developer) about different aspects
of Gluster triggered a memory of something a FOSDEM visitor asked: Would
it be possible to have a tiered Gluster volume with a RAM-disk as “hot”
tier? This is not something we can do in Gluster now, but it seems
that dm-cache can be configured like this. dm-cache just needs a
block-device, and that can be created at boot. With some config-options
it is possible to setup dm-cache as a write-through cache. This is
definitely something I need to check out and relay back to the guy
asking this question (he’s in the interesting situation where they can
fill up all the RAM slots in their server if they want).
Upstream testing the CentOS CI is available for many open source
projects. Gluster will be using this soon for regular distaf test runs,
and integration tests with other projects. NFS-Ganesha and Samba are
natural candidates for that, so I encouraged Michael and Guenter to
attend the CentOS CI talk by Brian Stinson.
Because the (partial) sysadmins for the Gluster infrastructure (Jenkins,
Gerrit, others servers and services) have too little time to maintain
everything, OSAS suggested to use the expertise of the CentOS team.
Many of the CentOS core members are very experienced in maintaining many
servers and services, the Gluster community could probably move much of
the infrastructure to the CentOS project and benefit from their
expertise. KB Singh sent an email with notes from a meeting about this topic to the gluster-infra list. It is up to the Gluster community
to accept their assistance and enjoy a more stable infrastructure.
Wow, did you really read this up to here?! Thanks 🙂
What a busy month this past month for Gluster! We’ve got updates from SCaLE, FOSDEM, our Developer Gatherings in Brno, DevConf, noteworthy threads from the mailing lists, and upcoming events. From SCaLE: Richard Wareing gave a talk at the Southern California Linux Expo about Scaling Gluster at Facebook. More at Scaling GlusterFS at Facebook From …Read more
Richard Wareing gave a phenomenal talk at Southern California Linux Expo on Saturday, January 23 about scaling GlusterFS at Facebook. In his own words: GlusterFS is an open-source (mostly) POSIX compliant distributed filesystem originally written by Gluster Inc and now maintained by RedHat Inc. Here at Facebook it had humble beginings: a single rack of …Read more
We’re kicking off an updated Monthly Newsletter, coming out mid-month. We’ll highlight special posts, news and noteworthy threads from the mailing lists, events, and other things that are important for the Gluster community. Community Survey Followup Our community survey results from November are out on the blog! News and Noteworthy Threads from the Mailing Lists …Read more
In November 2015, we did our annual Gluster Community Survey, and we had some great responses and turnout! We’ve taken some of the highlights and distilled them down for our overall community to review. Some interesting things: 68% of respondents have been using Gluster for less than 2 years. 3 shall be the number:The most …Read more
In my previous post, I talked about the sharding feature – what it does, where it is useful, etc. You can read that post here: http://blog.gluster.org/2015/12/introducing-shard-translator. When we designed and wrote sharding feature in GlusterFS, our focus had been single-writer-to-large-files use cases, chief among these being the virtual machine image store use-case. We are happy …Read more
GlusterFS-3.7.0 saw the release of sharding feature, among several others. The feature was tagged as “experimental” as it was still in the initial stages of development back then. Here is some introduction to the feature: Why shard translator? GlusterFS’ answer to very large files (those which can grow beyond a single brick) had never been …Read more
I have been working in GlusterFS for quite some time now. As you might know that GlusterFS is an open source distributed file-system. What differentiate Gluster from other distributed file-system is its scale-out nature, data access without metad…
Some quick highlights from our Gluster talks at LinuxCon EU, as well as slides available as a PDF below. Bitrot detection in GlusterFS – Gaurav Garg, Venky Shankar BitRot detection is a technique used to identify certain “insidious” type of disk errors where data is silently corrupted with no indication from the disk to the storage …Read more
Some quick highlights from our Gluster talks at LinuxCon EU, as well as slides available as a PDF below. Gluster Automatic File Replication Talk from Ravishankar Narayankutty: The talk started off with a quick-start introduction to the gluster lingo and its keywords, relating them to a block-diagram representation of the components. It then went on to …Read more
Sometimes the solutions we put in place turn out even better than what we originally hoped. That could sum of the experience of Belgian Internet Service Provider RIS Belgium,which turned to Gluster to solve the problem of distributed storage and ended up getting more benefit from the solution than they expected. Initially RIS, a web …Read more
There is a growing discussion in the IT world about the ways in which we, as information technologists, will approach managing the world of the small.
There are two aspects of current technology that fall into this category of “small”–containers and the Internet of Things. Both technologies were the subject of two intriguing keynotes at the opening session of All Things Open yesterday.
It’s been no secret that Intel is keenly interested in the Internet of Things (IoT) of late, and Intel’s Director of Embedded Software in the Open Source Technology Center Mark Skarpness laid out a concise look at how the hardware computer company is approaching the promise of a world of interconnected IP devices.
Currently, there are three obstacles that Skarpness perceives as a barrier to IoT growth: a lack of knowledge in embedded systems, a lack of knowledge and implementation of best security, and determining the most efficient ways in which IoT devices can communicate with the cloud and each other. This last aspect is particularly tricky; presently, many IoT devices operate in a siloed fashion. one or many sensored devices will talk to a control device, which in turn will talk to master server in the cloud that will collect data or deliver optimized operation instructions. But Company X’s devices will usually only work with Company X’s systems. Not Company Y or Z. This situation has led to what some wags on the press have called “The CompuServe of Things.”
This is clearly a situation about which Skarpness and his colleagues are well aware, and he did not try to give it short shrift when he told his audience what the path to better IoT would be. Issues like connectivity and customization of IoT devices in terms of form factor and purpose were two major elements of Intel’s IoT plan moving forward, Skarpness said, along with the aforementioned need to lock such devices down.
It was the fourth element of Intel’s plan that particularly caught my attention: building control systems that provide device management within the unique IoT ecosystem. Not only are deployed IoT devices found in very high numbers (think swarm instead of farm), but they can also be hardware-frozen. If you have a number of sensors buried in the road, for example, you can’t just dig them out and update their onboard hardware. You may not even be able to update the software; some devices may be transmit-only, not receive.
Managing devices in such circumstances can be tricky at best. The numbers alone can be daunting; sensor-equipped devices can range in the tens of thousands and the amount of data produced can create a huge firehose of incoming data. Indeed, many of these issues may seem familiar to those wrestling with another “new” bit of technology: containers.
Like IoT devices, containers are also proving to be a challange to manage. Containers and the microservices they host are not so much managed as orchestrated, according to the next keynote speaker, Sarah Novotny, Head of Developer Relations at NGINX.
Novotny referenced Skarpness’ outline of IoT concerns in her discussion of how information technologists should be approaching the future.of technology. Novotny’s talk highlighted Alan Gopnik’s 2011 New Yorker article “The Information,” which decribed how people then viewed the influence of the Internet on their lives: “call them the Never-Betters, the Better-Nevers, and the Ever-Wasers.” Novotny skillfully weaved these terms into the ways in which we view current technologies, including containers.
For containers, the “Better-Nevers” would look at the problem of managing containers and highlight the huge complexities in dealing with app-centric tools that in some cases, may only have a mayfly-like lifespan. The “Never-Betters” may see the positive benefits of microservices and realize that their benefits will outweigh the costs.
This may very well be true, but whatever your outlook, the challange of maintaining a cohesive form of management in container or IoT space is very real. As powerful as tools like oVirt, RDO, and ManageIQ may be, they are designed for managing whole machines (real or virtual), not potentially millions of containers or inerconnected hardware devices. This is why you see such emphasis placed on projects like Atomic, which are a new step towards a world where managing the small is even more of a critical need than managing the large.
This is a case of art (as technology) imitating life. Our tools are becoming reflections of our biology: where untold billions of microsystems make up the macrosystems we call the life we see around us. The transition won’t be easy, but the path to this kind of system deployment seems clear. It is fortunate the innovation that’s inherent in open source will help speed things along.