Bench Test - Xeon E5050 vs E5410

[Got2GoLV]

NOTE: I didn't think I had to clarify this, but this is strictly a comparison between A and B processors. Nothing more. This is not a post to show any particular performance of any specific hardware combination. other than same hardware, CPU A vs CPU B. Your hardware choices make a big impact in your performance. This is just to illustrate that point in regards to CPU choices.

I had some extra hardware laying around and decided to do some bench tests.

vSphere/ESX 5.5 box with 3 data stores served from the FreeNAS 9.2.1 box.
1- iSCSI (first column)
2- iSCSI (second column)
3- NFS (third column)

sync=always on all tests.

Testing was done from a Windows 2012 R2 VM running on the ESX box.

I basically benchmarked the top row, shut down the FreeNas box, changed out CPUs, booted FN, and then benchmarked the bottom row.

The ESX box and Windows VM were never revolted between tests/CPU change-out.

Top row is the Xeon E5050.
Bottom row is Xeon E5410.
Everything else is exactly the same hardware/software/settings between tests.

 

[cyberjock]

You do realize I can do specific tests that make it saturate 10Gb, or go so slow you'll want to shoot your eyes out? Benchmarks provide insight but are not to be used literally.

 

[Got2GoLV]

What did you think my point on this post was when you wrote this reply?

I'm not sure what your argument is.

This was just a test comparing results between 2 different CPU models. Nothing more, nothing less. Nothing implied or inferred other than the information shown.

If you assumed I meant something I did not explicitly state, then that's your assumtpion, and not my statement.

Relax, have a beer.

You do realize I can do specific tests that make it saturate 10Gb, or go so slow you'll want to shoot your eyes out? Benchmarks provide insight but are not to be used literally.

 

[jyavenard]

I find it very surprising it would make such a difference. Especially as both are rather powerful enough for this use

 

[cyberjock]

I find it very surprising it would make such a difference. Especially as both are rather powerful enough for this use

Are we even talking about the same CPUs here? The E5050 was discontinued Q1 2008. That means it's been "out of production" for 6 years! The E5410 is significantly better as it has a 1333FSB(over the E5050's 667FSB) and was released in 2007. Effectively the E5410 was released and then just a few months later the E5050 was discontinued. We're talking about night and day difference in CPUs. Totally different generations with totally different number of cores, cache size, etc.

What I mean is that technological differences in the CPU architecture can have a bigger impact if not optimized for that particular design. In particular, more L1/L2/L3 cache can have a major impact for a particular set of code that's just slightly too big for one CPU, but fits in the cache of another design. The throughput and latency when you have to go from RAM and not a cache is a night-and-day difference. It's like 3 orders of magnitude if I remember correctly. Not surprisingly, despite the 3Ghz CPU being the faster in clock speed, it is significantly slower in the above benchmarks than the E5410. The 4k benchmark clearly shows that at that point the CPU doesn't matter anymore as your pool is definitely the limiting factor. The QD32 gives you some indication of how far AHCI and NCQ/TCQ can potentially increase performance. But again, there's little information on how crystaldisk does it's test exactly, how that turns into transactions for iSCSI and NFS(they probably aren't the same), and then how ZFS handles it since ZFS is a COW filesystem and that has significant impacts for the benchmarking arena.

There's a thread by one of the FreeNAS developers using one of the higher end Haswells and it performs so fast it's actually having performance problems. Not sure if its a race condition that hasn't been seen before because CPU architecture marches forward or what. But it shows that engineering this stuff to be optimal is not a joke, and using benchmark results shouldn't be taken literally. I'm not saying we should take your results literally. Clearly your CPUs wouldn't end up on our "recommended" list solely because of their excessive power usage by today's standards. But clearly they may perform "adequately" in some situations.

And sorry if you took me the wrong way with the first post. I was simply trying to say that your benchmark values, while not horrible, provide a comparision between 2 CPUs(which you've reiterated), but the numbers as they stand don't really mean much except "the E5410 is faster than the E5050". But wasn't that an "obvious" analysis that didn't require analysis on your part? I'm somewhat confused I guess. Now if you benchmarked a bunch of the current E3-12XXv3s and provided values, that might blow my hair back to see the comparison. A performance versus cost comparison would be very handy for many people here.

 

[Got2GoLV]

And sorry if you took me the wrong way with the first post. I was simply trying to say that your benchmark values, while not horrible, provide a comparision between 2 CPUs(which you've reiterated), but the numbers as they stand don't really mean much except "the E5410 is faster than the E5050". But wasn't that an "obvious" analysis that didn't require analysis on your part? I'm somewhat confused I guess. Now if you benchmarked a bunch of the current E3-12XXv3s and provided values, that might blow my hair back to see the comparison. A performance versus cost comparison would be very handy for many people here.

I did not take your post the wrong way, it just seemed you did not comprehend the reason for my post and perhaps assumed I was boasting or somehow proving something with the actual transfer numbers.

The numbers do not matter in this. It was just a test to show the difference between A and B. In this case, 2 different processors.

And, was mainly to show how hardware choices impact performance.
I see a lot of people here that think that because FreeNAS is just a file server it doesn't need a lot of CPU/RAM and other hardware resources.

And I see a lot of arguments with users wondering why their box sucks in performance, or why they expected more out of their 10 year old hardware.

Well, instead of talking about it, I posted a direct A vs B comparison to help drive home the point.
I was not posting because I thought I had a breakthrough finding out one processor is faster than another one. Duh!

It was more direct at how hardware choices affect FN and similar filers. Thats it.

The post from jyavenard illustrates this.

 

[Got2GoLV]

I find it very surprising it would make such a difference. Especially as both are rather powerful enough for this use

This is why I posted. So others can see how big of a difference it can really make.

 

[cyberjock]

Ah. I did appreciate that comparison. I just take it for granted so often that I really misunderstood your reason for posting. I usually take the stance that if you've done your homework and realize you are very likely to want >8GB of RAM, your CPU is likely to be sufficiently powerful enough to accomplish your task with a reasonable performance. I own one of the crappiest Xeon 1366 CPUs that exist. I can't even find the part number, but it has no HT, is only 2.13Ghz, and only 4 cores. It's total crap. I thought I was making a smart choice as 4 cores without HT at 2.13Ghz ought to be enough for any fileserver, right?

It was completely crap in Windows Server 2008 R2.

But, when I moved the exact same hardware over to FreeNAS, it became a screamer. Throughput in excess of 1GB/sec were easily possible(if it weren't for the fact that I had 2x1Gb NICs). I'm fairly sure based on CPU usage I could have done almost 2GB/sec if I had owned a faster zpool.

Now, if the hardware is older than socket 1366, it is possible it's using FBDIMMs or other equally old RAM. The CPU will also be very inefficient in the power usage. For many people, the cost of running this old hardware is greater than buying a shiny new Haswell setup and running it for 3 years. As it is, I upgraded from my 1366 system to my Ivy Bridge setup last summer and I did the math and the hardware will pay for itself in just about 3 years.