Today I’ll be going over benchmark comparisons between Caliper’s new Dell Poweredge R620 servers and their older Poweredge R610 servers. I used Passmark Performacetest 8 for the benchmarking software. The CPU and memory benchmarks were run on a VM inside of vmware. The VM was the only active machine on the host and it was given the maximum amount of CPU and memory possible based on server specs. While that means this isn’t a ‘true’ benchmark of the R620 and R610 side by side it does compare them in Caliper’s real world usage examples.
Lets start with some basic server specs:
- Intel Xeon E5-2697 CPU
- 24 CPUs @ 2.7 Ghz
- 265GB memory
- 10GBE NICs
- Intel Xeon X5680 CPU
- 12 CPU @ 3.32 Ghz
- 196GB memory
- 1GBE NICs
First up is the CPU comparison. Overall the new E5 Xeon CPU had a much higher CPU mark score of 24812 vs the older Xeon X5680’s score of 12242. For what feels like a small model change this feels like a big jump in scores but it proves the point that its always better to upgrade equipment every 3-4 years, even if the model number change of the server doesn’t seem all that different. The improvements to CPU architecture have improved substantially in that timeframe. The only score to not have an almost 50% improvement was the single theaded benchmark with only a change from 1411 to 1628 in millions of operations per second.
Next up is memory benchmarking. The improvement to memory wasn’t nearly as across the board as it was for CPU. With an overall memory mark difference of 2258 for the R620 vs 1720 for the R610 most of the improvements were in read speeds, and then only in cached information. Write speeds didn’t see much of a substantial improvement nor did the latency. Threaded memory operations saw the biggest improvement at 53591 vs 15314 for MB transfered per second which can make an important improvement to some memory intensive applications but not all.
Finally we have the networking benchmarks. The networking tests were done over port 8080 with 90 seconds of data transfer per benchmark test. The first graph is just a quick average MB/s comparison between the two servers and the drastic improvement of 10 GB Ethernet over the older 1 GB Ethernet on the old server. While the R620 wasn’t able to reach its maximum throughput of 10,000MB/s it did have a 6x better average speed to other servers attached via 10GBE. “To client” refers to connections to a 1GBE standard PC over a small network. Strangely enough the R610 with its 1GBE did slightly better on the to client benchmarks, but this could also have been due to more overall network traffic during that benchmark run.
The last graph I have today is the networking benchmarks breakdown graph with both maximum and minimum networking values during the benchmark runs. This provides the best example of how the 10GBE network infrastructure still has room to grow while the 1GBE in the R610 server was pegged at its maximum for the entirety of the benchmarking tests.
While the model number doesn’t quite do it justice, the R620 is a fairly substantial upgrade over the R610 and in real world use at the Caliper office there has been a very noticeable improvement in application speeds. Not everyone may be ready to make the jump to 10GBE as the R620 doesn’t come with it out of the box but in Caliper’s case the change was well worth the performance improvements. The improved memory performance and ability to hold an extra ~70GB of ram per server were nice extra’s for a change out of a 4 year old R610 to a brand new R620. Keep upgrading your hardware, even if you feel it ‘runs well right now’ if it starts to fall out of your environments hardware lifespan goals!
As always the raw data can be found on my google docs here