Lefthand Rant

Warning: Rant content

So the talk of Lefthand math ties nicely into my topic for today. And yes, I’m going to rant. A number of years ago, we bought a SAN solution from Lefthand networks. It performed well. We were able to house Exchange and some file server stuff on it and it was just fine. We started down the path of virtualizing servers. We bought three more of the modules. At that point, they had discontinued the 150 model that we were used to. We smarted a bit from that, since we had negotiated future pricing on those units but we went ahead and bought 3 160 models.

The main difference between the 150 and 160 is the disk size. 4-500GB SATA disks vs. 4-250GB disks. We went along our merry way building virtual machines. It got very popular and one point, things just started to feel sluggish. We had the occasional machine just die and nothing seemed to quite be as snappy as it had been before.

We thought that perhaps it was a resource contention based on processor or bus throughput or something, so we bought two more at the same time we purchased Lefthand storage for a different capital hardware upgrade project. Things were good for a while and then slowed down again as we built more VMs.

After a lot of digging and researching, we theorized that we had simply run out of disk I/O available. We got Lefthand people to tell us what they thought an NSM160 could source I/O-wise. It was around 300 IOPS per module. We were pretty frustrated with the sales people and technicians just taking our money and not asking what we were using the storage for. They probably would’ve told us, as they did later, that you can expect to run 1-2 VMs per $8000 module. Crazy!

We were to blame too. We didn’t do our homework and we were just blissfully ignorant that I/O would’ve played such a huge role in our current mess. Here’s the rant:

As I mentioned in my last post, there are no, count’em no, tools available on the NSM160’s to monitor the I/O. They claim on some platforms that you can monitor via SNMP but that doesn’t work AND the counters that are available seem to merely be aggregate reads and writes since the last system reboot. If that data were available, you could write a script to capture the data and calculate deltas which would give you a IOPS over time number which you would then have to store somewhere and report on. They have announced some type of I/O monitor in the next release of their software this summer, but we’ve had the experience that they are not always timely.

I write this post to vent, of course, to also point out to do your due diligence. Even if a project starts out as a sort-of grassroots effort, research the gotchas of the project and consult with others in your industry to see what type of things they have encountered. If we had done that…well it’s water under the bridge now. Can’t dwell on it.

Explore posts in the same categories: Lefthand, iscsi, san, technology

Tags: iscsi, lefthand networks, san, snmp, technology

You can comment below, or link to this permanent URL from your own site.