Write speed benchmarks

[Helmonder]

There is a lot of talk going on about users wanting to know / benchmark / predict the transfer speeds of their unraid builds.

 

I propose we try and get some figures together in an easy way.

 

It would be important to all use the same tooling, I found a free windows based utility that works easy and quick, I am using lan speed test v1.1.7 (free windows utility, downloadable from http://www.totusoft.com/downloads.html ) with a Packet length of 1024MB.

 

For every test it would be important to know several variables, however the more variables the harder it would be to get the data combined and to compare and draw conclusions, I would therefor propose the following:

 

Variables:

 

- Type and speed of CPU

- Type of cache drive

- No plugins running (or not active, lets keep it simple again)

 

To be tested:

 

Writing:

 

- Directly to cache drive

- User share with cache enabled

- User share without cache enabled

 

Should this thread be succesful then I would be more then happy to compile the result (and keep it up to date) in an easy to read format (online and/or wiki).

[Helmonder]

See attached for the raw results (I will show in graphs including averages as soon as more results are in)

 

 

Unraid_Write_Speed_Benchmarks_1.txt

[Helmonder]

Tested system:

CPU: AMD Athlon 64 X2 Dual Core 4400+

Cache drive: Samsung SSD 830

 

 

WRITING             Time   Bytes p/second  Bits p/second Mbps

Directly to cache drive       12.2179253 83,811,283 670,490,264 670.4902640

User share with cache enabled 16.8459504 60,786,122 486,288,976 486.2889760

User share without cache enabled 36.6368691 27,949,987 223,599,896 223.5998960

[jowi]

Tested system:

Board/CPU:   Supermicro X9SCM-F / Intel i3-2120T / 4GB memory / unRAID v5.0 RC5

Cache drive: Corsair Performance Pro 128GB

 

WRITING             Time   Bytes p/second  Bits p/second Mbps

Directly to cache drive       11.0047163 93,051,013 744,408,104 744.4081040

User share with cache enabled 11.3585940 90,152,003 721,216,024 721.2160240

User share without cache enabled 24.7676458 41,344,261 330,754,088 330.7540880

 

Client system : Windows 7/64, Gigabit network.

Filesizes used: 1024MB/8192MB (same results)

[Helmonder]

Thanks !  I am anxious to see more results, this allready gives some nice info.

 

Our systems differ significantly in raw cpu power. Writing directly to the cache drive gives allmost the same performance, writing to the user share with or without cache show a big difference, this shows that cpu power has a significant performance advantage !  It also shows that in any case having a cache drive helps a lot in performance, the performance increase with a cache drive is even bigger with enough power under the hood.

 

Ofcourse we need way more data to get statistically reliable results.

[BRiT]

Any tests which transfer less than twice the physical ram of the unRAID system is useless.

 

Also for the info to be meaningful, the specs of both unRAID Server including memory and WinOS Client are needed.

[Helmonder]

You need to test a size that says something on the expected use, not on the physical capabilities of the system.. You are right that other factors add some effect to the test, however the major bottleneck in unraid is the parity calculation, that is not something we need to prove, its part of the specs of the system.

 

If you create 20 different variables to consider "important" you will never have a statistically large enough base to conclude stuff from..

 

What we need now (at least that is my opinion) is a large amount of tests to assess the major differences against each other, if we know that then it might be worthwhile to go a level deeper into more detail.. If you want to ofcourse, it is a matter of opinion if you want to spend time in assessing a 1 second difference on a 20 second job..

 

Primarily new users need info on deciding what system to build, they would not need info on the 1 second difference, they need info that goes towards the 10 second / 20 second difference, that's where the cpu is the largest contributing factor. Should you build on an Atom or an i3 ?  What effect does an i7 have ? etc. etc.

[UhClem]

Any tests which transfer less than twice the physical ram of the unRAID system is useless.

No. Any test which doesn't wait for its final write() to actually complete before "stopping its clock" is brain-dead.

 

All you do by making your test excessively long is reduce (but not eliminate) the inaccuracy (and waste testing time).

[Helmonder]

Thanks !  I am anxious to see more results, this allready gives some nice info.

 

Our systems differ significantly in raw cpu power. Writing directly to the cache drive gives allmost the same performance, writing to the user share with or without cache show a big difference, this shows that cpu power has a significant performance advantage !  It also shows that in any case having a cache drive helps a lot in performance, the performance increase with a cache drive is even bigger with enough power under the hood.

 

Ofcourse we need way more data to get statistically reliable results.

 

My results are very comparable to yours.. Really shows that having more processor power increases speed in using user shares..

 

I think it is great that unraid can run on a low profile system, that is the way I tried it out (with some old hardware), it performed great, stable as a rock and the speed was actually ok.

 

However when you want to buy new hardware it is a wise move to not go the low profile route.. Price difference is not THAT big and you get a lot back for it... I also notice that web-frontend response is quicker, more snappier. It worked fine on my Athlon X2 but speed increase is absolutely noticeable using the i3 2120T. Glad I made this choice!

[napalmd]

Asus p4p800 / intel pentium 4 3,0Ghz / 1Gb ram / Unraid 4.7

 

No cache drive

 

                        ---Writing---    ---Reading---

Packet length  :      1,024,000,000    1,024,000,000

Time to complete:      101.0540684      32.1914923

Bytes per second:      10,133,189        31,809,647

Bits per second :      81,065,512        254,477,176                     

-------------    -------------

Mbps:                  81.0655120        254.4771760

 

Too slow right?

Will this speed improve by upgrading to core i3?

[napalmd]

Yes they are spun up, made the test after a reboot of the server.

The slow of it is being discussed here http://lime-technology.com/forum/index.php?topic=21960.new;topicseen#new

[Helmonder]

Tested system:

CPU: Intel I3 2120T

Cache drive: Samsung SSD 830

 

 

WRITING             Time   Bytes p/second  Bits p/second Mbps

Directly to cache drive       11.1383789 91,934,384 735,475,072   735.4750720

User share with cache enabled 10.4695345 97,807,596 782,460,768 782.4607680

User share without cache enabled 29.3694526 34,866,159 278,929,272 278.9292720

 

[Helmonder]

I would like to have 10's of tests more for this to be somewhat representable but based on the current available results some stuff is visible:

 

1) First (obvious ofcourse) result is that hardware does matter, different hardware gives definite different results in performance.

 

2) A cache drives has a definitive positive effect, no matter what the underlying hardware is, writing to a cached enabled share is twice as fast, so (taking only speed into account) having a cache drive is a good idea;

 

3) The differences with hardwarebecome apparent as soon as unraid itself is thrown into the mix, if you write directly to a cache drive, and therefor bypass the unraid system itself, then speeds are approximately the same no matter what the underlying hardware is.

 

I would be very interested to see the results of someone having an i5 or i7 processor, I would suspect that those will not lead to substantial increase of speed.

 

I would also be interested in more results with different types of cache drive and processors. An Atom D525 or 515 would also be good for comparison.

 

With enough base info I think we can work towards a build advice for new builds.

 

My feeling is that it is great that unraid runs on the stuff you still have lying around, this is a great way to experience the product and find out, but at some point you might want to buy new hardware and then some idea on the most suited hardware would be nice.

[Helmonder]

Asus p4p800 / intel pentium 4 3,0Ghz / 1Gb ram / Unraid 4.7

 

No cache drive

 

                        ---Writing---    ---Reading---

Packet length  :      1,024,000,000    1,024,000,000

Time to complete:      101.0540684      32.1914923

Bytes per second:      10,133,189        31,809,647

Bits per second :      81,065,512        254,477,176                     

-------------    -------------

Mbps:                  81.0655120        254.4771760

 

Too slow right?

Will this speed improve by upgrading to core i3?

 

Based on what the data shows right now (though we have limited measuring points and are not statistically sound yet) youy're first move would be to start using some kind of cache drive, this will give you the largest increase in write speed (take into account if the disadvantages of using a cache drive are acceptable to you).

 

Upping the processor will also help but not as much as the cache drive. It is also a more intensive move so I would first do the cache drive.

 

Please note that the above is valid for a system that is set up "right", if your speed is hampered because of something specific to your configuration then that should be solved first (ofcourse changing motherboard and cpu will also fix that most probably ;-)