HPE Discover 2016 - Day 3

Goodbye HPE Discover 2016

Day 3 was just as packed as days 1 and 2.  The highlights of Day 3 for me were more CDA, a meeting with the VP of the HPE Central Region, Sitescope monitoring of CloudSystem, NVDIMM performance testing, and composable programming.  Here are the previous posts from Day 0, Day 1 and Day 2.

8:00 AM Confidentiality Disclosure Agreement Area

As with yesterday, I am obligated by the CDA that I signed not to divulge details of future developments.  What I will say is that I was impressed by what I saw in a future SMB OS and extremely impressed with where NVDIMMs are going.  Non-volatile memory, along with photonics, is key to the high-speed promise of The Machine. I saw a demonstration of database performance on a small server that matches the performance of HP-UX Superdome 2 servers.  Later in the day, I attended a non-CDA HOL on NVDIMM where we ran database performance tests.  Keep reading for those public results.

8:30 AM Meeting with Andrew Manners, VP HPE Central Region, and Mike Warbiany, HPE Director of Channels Central Region.

Not much of interest here to non-Rolta AdvizeX employees.  Basically, Rolta AdvizeX has been a great HPE partner for over 30 years and there are business segments that HPE would like to work more closely with Rolta AdvizeX on growing.

9:00 AM Monitoring the health of HPE Cloud Service Architecture in your Datacenter (Hands-on Lab)

Having never used HPE Sitescope before, it was nice to see how powerful SiteScope can be, even without agents.  It was also fun to look behind the curtains and play with the HPE Support, internal-only, tool LabRAT.  LabRAT is used by HPE support engineers to request and build duplicate customer environments for testing.   Basically, in this lab we:
  • Logged in to LabRAT
  • Used LabRAT to automate the creation of resources in CloudSystem
  • Logged in to SiteScope
  • Created a monitoring group for our resources
  • Imported monitoring templates for our resources
  • Started monitoring resources and running reports
SiteScope is just one of the tools you get with Cloud Service Automation (CSA) that is built into CloudSystem.  Other significant tools included in CSA are Operations Orchestration and Server Automation.  The often asked question with SiteScope and Server Automation is "When would I just use OneView?".  OneView is and HPE-specific management platform that began with managing and automating HPE servers, networking and storage and is slowly branching out to non-HPE devices.  SiteScope is vendor-neutral and uses open standards for monitoring.

You can grab a copy of the lab guide here.

11:30 AM Composable Infrastructure ecosystem in action; HPE Helion CloudSystem (Hands-on Lab)

As we well know in the presales profession, live demos are thrilling because, at any time, they could stop working.  The networking in the hands-on lab floor was a bit cranky at times during the week.  Unfortunately, the network decided that the CloudSystem lab was not going to happen at 11:30.  The instructors apologized, I checked the other available HOLs for 11:30 and happily trotted off to the NVDIMM lab.

11:30 AM Deploying HPE 8GB NVDIMMs on ProLiant Gen9 servers running Linux and Oracle (Hands-on Lab)

Ground-breaking.  No, really.  

I am a creative person and IT is an extremely creative field.  I sometimes say that I have had the same job my entire career - explaining how the latest advancements in IT produce positive business benefits.  Due to the creativity of engineers in IT, every day is different.  Non-Volatile Dual In-line Memory Module (NVDIMM) is a groundbreaking technology in terms of the speed of application transactions.  There are healthcare, business and engineering applications where processing data 1,000 times faster is required.  Think about modeling drug formulations against disease or analyzing security attacks in real time as two examples. 

This was a fascinating lab.  Each station had access to a Linux server with NVDIMMs installed.  We started by looking at the iLO options for NVDIMMs.  The HPE Smart Storage Battery enables using NVDIMMs are read/write.  If the battery is missing or degraded, the NVDIMMs switch to read only.  We also looked at firmware.  Each NVDIMM has a controller with firmware embedded.  If you pause the boot of the server to get into setup, you can set NVDIMM options in System Configuration --> (RBSU) --> System Options --> Memory Operations --> NVDIMM Memory Options:
You will see that you have the option to erase the persistent memory if needed.

Once you boot into Linux, the ls /dev/pmem* command will display the NVDIMM devices in the system and the fdisk -l /dev/pmem* command will show the NVDIMMs as "disks".

The next step was to use the RESTful api to set memory interleaving on the NVDIMMs from Linux.  The hprest.rpm package is needed for this.  Once the hprest.rpm is installed, interleaving can be turned on and committed.

We then ran the same IO performance test against the NVDIMM "disks" as 1) disks, 2) as memory non-direct access, and 3) memory direct access.

Disk access results:

Non-direct memory access results:

You will notice that CPU use goes up significantly, but that IOPS increases 1,000X! If you have been in the IOPS field for a while, you know that getting 258,536 IOPS from a single server is unheard of.

Direct memory access results:

You will notice that the overall CPU utilization is similar to non-direct memory access, but that the IOPS increased by 14.5%.  Also nice to note the balance of read and write performance.  That is not always the case with SSD drives.

In the lab guide, you will find other comparisons for Oracle redo logs and OLTP benchmarks.

You can grab a copy of the lab guide here.

2:00 PM Programming the Composable Infrastructure (Hands-on Lab)

HPE has been investing R&D and marketing dollars in "Composable Infrastructure", hardware that is as flexible and programmatic as software.  OneView is the single tool to manage and automate this composable infrastructure.  The OneView HTML5 interface is not the only method of taking advantage of OneView capabilities.  OneView has a full RESTful interface and can be automated using many DevOps tools as well. 

A nice tool to use to to test any RESTful API is the POSTman Chrome extension.  You can install POSTman here

Here is the output of a simple "GET version" REST request of OneView

You will see that OneView is version 3.00

Many REST actions will require authentication.  Once authenticated, a session token can be used for authentication for the life of the session.  To get the session token, POST userName and password values. The result will be a sessionID value.

Now that we have our sessionID, we can start to drill down into the inventory that OneView knows about.  The top level is data centers.  Sending a GET to the datacenters URL will return JSON of the inventory at the data center level.

You can keep GET-ing URLs of different objects to see their children.

The next step in the lab is performing the same REST drill down, but in JAVA.  You can read about that in the Lab Guide if you are interested.

The next step in the lab is using the extremely powerful Powershell OneView module.  Powershell is my favorite method of automating OneView tasks.  One of my peers wrote a Powershell script that initializes 32 blades, 4 FlexFabric modules and two 3PAR arrays in 20 minutes.  You can find the OneView Powershell module on GitHub here.

To connect to OneView, the PowerShell command looks like this:

Connect-HPOVMgmt -Appliance oneview4prog.etss.lab -UserName administrator -password password

To list the servers in your environemnt, type this command:

You can download the Lab Guide for this lab here.

That about wrapped up the final day for me and another educational and successful HPE Discover.  I hope you have enjoyed this blog post series and I welcome your feedback.