NAS review performance testing with port trunking

[jprule]

Hi all,
I am new here and now have a question about the way NAS machine performance is reviewed by smallnetbuilder.

I have been checking the NAS chart for performance of some boxes my company is considering buying.

And found that although most of the machines come with two GbE ports and support trunking, their performance was tested using only one port.

I really don't understand why and expect for your explanation. Moreover, how am i supposed to know the max. performance of a NAS based on the 1-port performance data?

Hope for your help. Thanks.

 

[GregN]

I really don't understand why and expect for your explanation. Moreover, how am i supposed to know the max. performance of a NAS based on the 1-port performance data?

Hope for your help. Thanks.

I can't speak for Tim, the man who is SNB and runs the performance charts, I'm sure he'll explain better than I can, but to take a stab at it....

Most trunking requires equipment that supports 802.3ad/ax, like a managed/smart switch, which is unusual for most users here, additionally link bonding, or aggregation only works with multi-sessions (you don't get twice the bandwidth to any single session ). With that, and that running full out on a single gigabit connection that, for example the NASPT tests achieve, will tell you what performance you'll see as a client, and is overall very telling ( I can't see the multithreaded Office NASPT test seeing much improvement with bonding ).

Given that most folks don't shop for aggregated performance, and the full extent of the performance is usually plumbed, it just doesn't make much difference.

 

[stevech]

Might I add: reaching the capacity of ONE gigabit LAN port for disk transfers is a teeny bit of a challenge for smallNetBuilder-class NASes.

 

[thiggins]

Both Greg and Steve's answers are pretty much spot on. Aggregated NAS ports only help in multiple client situations. Transfers from a single client, even if it has aggregated ports use only one port per transfer and the test transfers are not multi-threaded.

 

[jprule]

Thanks guys for all your replies. So let me summarize the reasons here:
1. SNB used the single-threaded testing method, and since transfers from a single client use only one port, it makes no sense to apply link aggregation.
2. The single-threaded method is used because (1) fully-functional port trunking requires networking equipment supporting 802.3ad/ax, which is not common for SNB users (2) it tells the performance user can see as a client (3) due to CPU power limitation, it's difficult for the NASes tested by SNB to exceed the 1 gigabit LAN threshold.

Do i understand you all right?

 

[GregN]

1. SNB used the single-threaded testing method, and since transfers from a single client use only one port, it makes no sense to apply link aggregation.

Single client, the NASPT tests are multithreaded. But no one client/node can exceed 1G.

Router 1G-----\ LACP ( 802.3ad )
Laptop 1G----- >Switch ==2G==NAS ( No one link gets over 1G to NAS )
Server 1G-----/

2. The single-threaded method is used because (1) fully-functional port trunking requires networking equipment supporting 802.3ad/ax, which is not common for SNB users

Results would be meaningful only to a very small group of the readership. And the current results give a very capable picture of expected performance.

(2) it tells the performance user can see as a client (3) due to CPU power limitation, it's difficult for the NASes tested by SNB to exceed the 1 gigabit LAN threshold.

The results you see are, I think, the best indicators you can gather to characterize that NAS products performance.

Take a look at the iSCSI results, which have significantly less overhead, both bandwidth and CPU (block level, native filesystem, no samba) - they are not dramatically different. Gig Bandwidth is but one part of the overall performance equation. Look at it this way, when moving from 100Mb to 1000mb you don't see a tenfold increase in the client performance with the dramatic increase in bandwidth, do you? The pipesize is only part of a much bigger melange of factors.

Some other factors that would impact performance that are not measured are drive speed, spindle size, stripe size, cache policy ( Write Through or Write Back ), and primary NAS tasking (Media vs Backing up vs Generalized storage ).

Do i understand you all right?

Not sure. Hopefully this helps.