HPE Nimble Storage Tech Blog
cancel
Showing results for 
Search instead for 
Did you mean: 

Experiences upgrading from NimbleOS 1.4.9 to 2.2.5 with VMware NCM

joebrug52

Just wanted to share my experiences/results with the community. Huge thanks to Nick Dyer for the guidance during our first Nimble OS upgrade!

Setup: CS260 with 1GB iSCSI links, 3 VMware esxi 5.5 hosts.

We've been on OS 1.4.9 for a long time. When the system was configured, our SE used some scripts to set up the multi-pathing to work with VMware using round robin, etc.. this seemed to be working fine, didn't know any better.. but once we upgraded to NOS 2.x using VMware NCM, our IOAnalyzer tests were drastically different!

First, the upgrade: Downloaded 2.1.8 and hit Update, it took probably about 30 minutes or so, but it worked flawlessly. No downtime noticed by anyone, except for me, who had some constant pings set up for the controllers to see when they rebooted. I was about to upgrade immediately to 2.2.5, but noticed that the Hot Spare started reconstructing itself. Found out that this is due to the Raid-3P introduced in Nimble OS 2.x. Was told I could probably go ahead and upgrade to the 2.2.5 OS, but decided to wait until the reconstruct was finished. When that was done (roughly 24 hours), I went ahead and did the upgrade to 2.2.5 during business hours. Again, no one noticed a thing, and it went very smoothly and quick (Much faster than 1.x to 2.x).

After upgrading, I downloaded I/O Analyzer – VMware Labs

I picked a couple of the tests, fairly randomly and compared the results. The tests only ran for 120s (default). The first two tests were from the current setup: OS 2.2.5 without VMware NCM/Automatic Connection mode/Re-balancing. The last two tests were WITH NCM/Automatic/Re-balancing installed/configured/enabled. You'll note quite the difference!

IOAnalyzer.JPG

Screenshot of Nimble GUI during Read test without NCM (still using "scripts" from 1.4.x):

NimbleReadwoNCM.JPG

Screenshot of Nimble GUI during Read test WITH NCM+Automatic Connection Method+Re-balancing installed/enabled/configured:

NimbleReadwNCM.JPG

The Nimble/VMware setup seems much happier using the NCM with OS 2.x. Obviously, something wasn't configured properly, or things have just gotten that much better with 2.x+NCM. Also, the 4 Ethernet interfaces of the nimble went from maxing out at around 50MB/s to about 120MB/s! Evenly distributed across all 4 interfaces. Much more throughput, and way more IOPS!

About the Author

joebrug52

Comments
rfenton4

Thanks for sharing Joe - there was a significant change in 1.4 to 2.0 (we added Scaleout) which explains why the process was probably a little longer thaN the in family 2.x releases

jpmeredith

Wait about 2 weeks and check your cache utilization in Infosight -- I bet you'll be pleasantly surprised again! We saw significant performance gains after the 2.x upgrade.

joebrug52

You mean it'll stop telling me it recommends an upgrade to cache?

ndyer39

Nice work Joe!

The upgrade from 1.4 -> 2.x is not a trivial one on the underlying code as Rich Fenton mentions; we brought in Scale Out but also the ability to have intelligent multipath awareness for VMware and Windows. It's great to see the real-world results of said MPIO intelligence

On the whole, 2.2.5.0 is a great release which seems to be very stable in the field, and with our new flash caching algorithm it's possible to realise even better cache hit performance than what was needed previously.

jpmeredith

Mine did. Hovering around the 50-60% mark, and cache hit rates on the Volume list page are better than ever.

thegrove52

Mine also is doing much better.  In addition to it no longer telling me to upgrade the cache my CPU utilization went from a solid 500% where it had been running since November down to below 100%.