In our cloud, it seems like there are two things I can never have too much of … that being compute power and bandwidth. As soon as I stand up a chassis with blades in it, a workload occupies that space. The same is with our bandwidth … as soon as we make more bandwidth available, some form of production or replication traffic scoops it up.
Introduction: Silver Peak
About 6 months ago, I started to look at Silver Peak and their VX software. I have a lot of experience in the WAN optimization area and the first couple of concerns that came to mind were ease of configuration and impact to our production environment in that what type of outages would I have to incur to deploy the solution. I did not want to have to install plug-ins for specific traffic and if I make changes, what does that do to my firewall rules?
After reviewing the product guide, I spoke with the local engineer. The technical pitch of the software seemed fairly easy:
1) Solves Capacity (Bandwidth) by Dedupe and Compression
2) Solves Latency by Acceleration
3) Solves Packet Loss by enabling Network Integrity
The point to Silver Peak is they are an “out of path” solution. I migrate what I want when I want. There are two versions of their optimization solution available. The first is the VX software that offers multi-gigabit using a virtual appliance. The second is the NX appliance, which is a physical appliance that offers up to 5Gbps throughput. Since we are a cloud provider, no stretch to guess that I tried the VX virtual appliance first.
The Deployment
After deciding what we wanted to test first, it was on to the deployment and configuration. Another cloud engineer and I decided to install and configure on our own with no assistance from technical support. I say this because I wanted to get a real sense of how difficult it was to get the virtual appliance up and running.
(By the way, technical support is more than willing to assist and Silver Peak is one of the few vendors I know that will actually give you 24×7 real technical support on their product during the evaluation period—it isn’t a “let’s call the local SE and have him call product support”, but true support if it’s needed.)
After our initial installation, turns out we did need a little support assistance, because we didn’t appear to be getting the dedupe and compression rates that I was expecting. Turns out, after the OVA file is imported to the virtual environment, I should have also set up a Network Memory disk—this is what allows the appliance to cache and provides the dedupe and compression. Since I didn’t have this turned on, the virtual server memory was used … my fault. Even with the support call, we had the virtual appliances installed and configured within 1.5 hours. If I take out the support call, I can literally have the appliances downloaded, installed and configured within an hour.
If I take out the support call, I can literally have the appliances downloaded, installed and configured within an hour.
What We Were Optimizing
Scenario 1
We had two scenarios we were looking at with our evaluations. The first scenario was our own backbone replication infrastructure. We replicate everything from SAN Frames, backup data and archive sets. Our test encompassed two of our datacenters and moving workloads between Chicago and Denver. My first volume was a 32TB NAS volume. Since this solution was “out of path”, I simply added a route on our filers at both ends to send traffic to the local Silver Peak appliance. When Silver Peak is first configured, a UDP tunnel is created between both appliances. Why UDP? Because UDP gets through most firewalls without very complex configurations. When the appliance receives traffic destined for the other end, the packets are optimized and forwarded along.
With this test scenario, we saw up to 8x packet reduction on the wire. That being said, we were able to replicate about 32TB of data between filers in just over 19 hours.
Scenario 2
Our second scenario was replication from a remote customer site using a 100Mbps Internet circuit connecting via a site-to-site VPN. Latency on a good day was 30ms with spikes to 100ms.
In trying to replicate the customers SAN data between their site and ours, stability was a huge issue. Our packet loss on the line was very high. High to the point that replication would stop on some of the larger datasets.
We installed the VX5000 between sites to optimize the WAN traffic. What we found was line utilization went to 100% with packet loss almost to 0%. Our packet reduction was constantly around 4x. While the customer was initially hesitant to try the appliance installation, they too were surprised how quick and simple the installation was.
While the customer was initially hesitant to try the appliance installation, they too were surprised how quick and simple the installation was.
With the Silver Peak appliance, we changed out daily replication window from 15+ hours to less than 2 hours.
Conclusion
Surprises in the datacenter are never a good thing. However, being in this space for over 20 years, I have experienced my share of unexpected surprises (not many of them good.) In both our test scenarios, I can honestly say, I was surprised with Silver Peak … and in a good way. I was surprised how easy the installation and configuration was. I was surprised how well their solution worked. I was surprised how good their sales and engineering teams really are.
Photo credit: Emma and Kunley via Flickr