3

We replaced our main line of business server with a new one. The old server had dual Xeon X5690s in it (2 generation old chip), the new server has quad Xeon E7-4807s. We expect processor performance to at least stay the same.

Our main java app showed a 50% performance hit and on running a sysbench against the two machines, the old processors were 50% faster.

Is this a particularly bad chip ? Is there something we are missing performance wise ?

(Os on both is Centos 5.10 x86_64 and memory is 128GB)

ewwhite
  • 197,159
  • 92
  • 443
  • 809
David
  • 33
  • 3
  • 1
    Is your app able to take advantage of multiple cores? – Grant Apr 10 '14 at 21:56
  • 1
    Older chip. Take a look at the [E5-4607 v2](http://ark.intel.com/products/75794/Intel-Xeon-Processor-E5-4607-v2-15M-Cache-2_60-GHz) for an example of something similar but recent. – Brian Apr 10 '14 at 23:51

2 Answers2

7

Kinda...

This is a good lesson in resource-planning and making hardware decisions across CPU lines and generations. I've had the unique experience of needing high-end CPUs for applications in a few different industries, but also the luxury of being able to test and benchmark before major architecture changes.

As a result of this, my rule is to examine the product lines and move to processors that have comparable positions in the portfolio when you make hardware changes.

What's better? A 2010 Mercedes S-Class or a 2014 Mercedes C-Class? It depends...

  • The Intel Westmere x5690 was a 3.47GHz hex-core processor. It was Intel's top-end mainstream server CPU during its product lifetime.
  • The Intel Westmere-EX E7-4807 was a 1.86GHz hex-core processor meant for quad-socket boards. This CPU was at the bottom of the product line. It was released in the same quarter as the x5690.

Intel Comparison Chart - Comparing the x5690 and E7-4807

So what's wrong?

Your "new" server is also running an older architecture. Remember, the bigger server platforms (quad-socket, etc.) don't rev as often as the smaller systems. You have a CPU that's running around half the clock speed of the previous CPU, but of the same generation, thus the performance drop. The E7-4807 is also missing Turbo Boost.

I'd be interested to know what prompted the upgrade. You likely had the raw CPU performance you needed. Was it memory or another factor? What was the bottleneck?

A better move if you simply needed additional cores would have been a newer dual-socket system, but with top-line Intel E5-2643v2 (3.5GHz) or E5-2697v2 (2.9GHz) CPUs.

ewwhite
  • 197,159
  • 92
  • 443
  • 809
  • 3
    This sounds more like a _downgrade_ than an upgrade. – Michael Hampton Apr 10 '14 at 23:30
  • 2
    @MichaelHampton Yes, a big downgrade. That's why I'm curious how this server was chosen. – ewwhite Apr 11 '14 at 02:41
  • Was trying to optimize database performance, moved to 2u Quad processor mostly to get space in the chassis for a better disk subsystem - decent RAID controller running SSDs so I could put the database and logs on different disks. This did work - db performance is greater than 2x improved, but had the unintended consequence for tanking main app performance. Turns out the CPU was far more important performance wise than the db... – David Apr 11 '14 at 12:49
  • 1
    @David well, it's possible to acquire a 2U 2-socket server with the disk expansion you need. Good luck in your efforts! – ewwhite Apr 11 '14 at 13:11
1

Your "new" server is also using 2 generation old processors. They are a newer, lower power iteration of that generation.

Unless you're intimately familiar with processor architecture, it's a good idea to stay within the same family and TDP when you're upgrading. Intel's nomenclature isn't always as straight forward as it could be, but the current replacement for the X5600 series is the E5-2600V2 series.

Here's a comparison of the processors in your old server, your new server, and the current replacement for your old server.

Jason
  • 718
  • 5
  • 15