Microsoft Ignite 2015 Thursday

Microsoft System Center Virtual Machine Manager: Technical Overview and Roadmap

Here are the highlights from this session:

– Now delivering new functionality as part of update rollups, not only fixes as was the case in the past
– Better release quality for update rollups (bravo!)
– Fully automated upgrade process from Windows Server 2012 R2 to Windows Server 2016
– Automated deployment of SOFS clusters
– Storage Spaces management integrated in the tool
– vSphere 5.5 core support is available, version 6 is coming
– A demo of the Amazon EC2 plugin for SCVMM was shown
– Better support for Azure is coming, right now very basic (you see VMs and you can start and stop, connect via RDP but that’s about it)
– They are using two main channels to gather feedback, support and UserVoice to figure out what they will work on next

Advanced Modeling and Calculations using the Power BI Designer

This was a good session to give me a feel of the usefulness of Power BI. During this session I was able to see how the Power BI designer tool was used to create DAX expressions to overcome common business problems. For instance, it often happens that you want to compare this year data in comparison to last year. It turns out it’s pretty straightforward to do this in DAX and there is a builtin function to do this (I think the function name is called SAMEPERIODLASTYEAR). Also covered in the presentation were new functions that are not yet available but are coming with an upcoming build of Power BI. Overall the tool looks very powerful and easy to use and is definitely something that I will investigate further once I get back the office.

Azure IaaS for Azure Stack

This gave a nice overview of what Azure Stack is for IaaS. In a short summary, it’s pretty much means that you’re running Azure on premise. If you learn to build automations using Azure PowerShell cmdlets, you will be able to use the same exact scripts to automate infrastructure provisioning on premise. They also took some time to explain how you can create template written in JSON to automate the deployment of services like a SharePoint farm for instance. The template format is quite flexible (you can call templates from within templates) and can easily define parameters.

Enabling Private Cloud Storage Using Servers with Local Disks

This session was probably one of the main reason why I wanted to come to Ignite as this was something I have been quite interested in since I first heard about it in TechEd Europe last year.

The technology while simple is quite powerful. They basically created virtual storage bus that spans multiple hosts in a cluster. Then the drives are presented to Storage Spaces as usual. You need at a minimum 4 hosts to achieve this (possibly 2 cluster in a box would do the job as they have 2 servers in each). Data is then spread across the nodes in the cluster in order to maintain redundancy depending on how many copies you have selected when you created your Storage Spaces virtual disk. One additional thing that I found interesting is that this requires ReFS instead of NTFS. The recommendation by Microsoft to move to ReFS is pretty clear as a number of session and new features in Windows Server 2016 requires this. This will make the solution more resilient as ReFS has quite a few reliability features.

There are quite a few benefits you get from that configuration:

– No need to physically collocate servers in the same rack or do crazy wiring job to have a fault domain per rack
– You can use pretty much use any type of disks (SATA, SAS, NVME) which gives you great flexibility and the opportunity to reduce cost
– You can run either in a hyper converged scenario or by exposing the storage through SOFS
– Capacity/data is rebalanced as nodes are being added and removed from the cluster
– In the event of a failure, the data is repaired automatically using spare capacity on other disks/servers
– If the server is restarted or shutdown for some reason, it will resync automatically with the remaining nodes in the cluster

The speaker also mentioned that it is looking at IO queue depth on each host to figure out where to read data from. We saw during one demo how the solution was able to deliver about 1.1M 4K IOPS at 1ms latency in average. The current maximum of nodes tested by Microsoft is 12 and they are looking at increasing that number before Window Server 2016 RTMs. The size of storage pools was also greatly enhanced in Windows Server 2016, going from 80 disks to 240 disks with an RTM target higher than that. RDMA again will be of great importance. In one demo, they compared the performance with both RDMA enabled and disabled. The difference was almost 50% less when RDMA was disabled. We saw the IOPS cutting to about 600K IOPS and IO latency go up to 2ms.

The group is also working with the Cluster Aware Updating group to make sure the patching process is right. As the server need to resync data after they are restarted, you don’t want to restart servers too quickly as this might create data loss/corruption issues.

Monitoring of the storage subsystem was also greatly enhanced in Windows Server 2016. The goal is to give more meaningful and actionable alerts to the user. For instance, if there a disk failure, this is surfaced only as a warning as this is not a threat to the overall cluster as everything keeps functioning just fine in that event.

Exploring Storage Replica in Windows Server vNext

This was a really great presentation by Ned Pyle about Storage Replica in Windows Server 2016. Storage Replica is a new feature that allows you to replicate volume at the block level. There’s quite a bit of flexibility in what you can do with the technology and how you can configure it. The technology supports both synchronous and asynchronous replication. With synchronous replication, you need to aim for latency below 5ms for this work properly, otherwise the performance of the system will become very slow as each IO done needs to be hardened on the remote log disk before being acknowledged to the initiating application. This means in practice that you are limited to about 50km. If you opt for async, there are no constraint on the latency and therefore the distance. In both cases you need to make sure that your log volumes are on super fast storage as this is the place where the IO will happen first. From the current performance test, there is about a 15% performance hit when you enable replication. Microsoft is looking at reducing that below 10% when the feature ships. There were also initially a performance hit in regards to encryption in SMB3 which should now be mostly eliminated.

The whole process of setting that up was greatly improved from the initial technical preview build from last fall. It’s pretty straightforward to validate if SR is an option in your environment and then set it up. We will definitely take advantage of this in a few scenarios!

Ignite Party

As everything at Ignite this year, you had to wait in line forever to achieve anything. The principle of inviting local restaurants to cater was a good one but maybe a bit impractical for such as large number of people attending the party.

We then had the chance to a DJ set by Lani Love which was then followed by an hour concert from Fall Out Boy. Both gave really good performances even though the crowd was quite eclectic. FOB even did an awesome cover of Beat It which I enjoyed quite a bit. Here a link to a video from someone at the even.


Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s