Intel SS4200-E speed

[ganryu-island]

Following is Read/Write test regarding SS4200-E(Latest firm & 1GB Cache).
The end of a html page.
http://www.ganryu.org/DS_pf_Check/

Soon, I'm planning to test by the next condition.
2GB cache
Intel Celeron 440(2GHz)

Ganryu

 

[Dennis Wood]

I can see one problem here in that you're testing 1GB files but your NAS also has 1GB of RAM. Ideally, your test files should be at least 4GB so that you are well outside of the cache values for the test workstation and Intel box. Right now you're measuring how fast files can be transferred to the NAS RAM...which does not reflect actual RAID5 read/write performance.

Our measured test values using a 5.3 GB test file set to the SS4200 are in the area of 28MB/s write, and 45MB/s (SMB2) read using a RAID 5 config on the NAS. RAID 10 on this box got us results at 34MB/s write and 39MB/s read. We're using the same firmware, but haven't touched the NAS RAM.

 

[ganryu-island]

Hello Dennis Wood,

Thank you for your advise.

I point out that
1000M bytes of file is being written in 9 times continuously, so almost all cache is always fully.

The next Network Link Speed chart is a copy test result of the 4GB file.
http://www.ganryu.org/ss4200e/2008-08-26_145659n.jpg
The read/write max speed of the ST336753LC is about 70MB/sec, so it seems dependent on this numerical value.

Regards,
Ganryu

 

[Dennis Wood]

Is it the same 1000MB file? Also, because Tim is using Iozone, and we've found it very accurate here, why not give it a try? Then your results will be more easily compared to the tests that Tim is doing here. The learning curve is a bit higher, but I'd be glad to help out if I can

Your results will vary from our numbers as we are using a 5.3 GB file test, as well as ffmpeg encode and calculating rates based on the file operation/time. In other words, it's a "real" test.

We've seen nothing here, including a 3 Drive workstation based RAID5, that can be written to at more than 45MB/s in RAID 5, and that includes the QNAP unit with a 5 drive array. Again, these are measured tests, not benchmark programs but what we do find is that Iozone (out of all the programs I tested) is an excellent predictor of real world performance.

 

[ganryu-island]

Is it the same 1000MB file?
No, I used the 4GB file made in mkfile command in Mac_OSX.
http://www.ganryu.org/ss4200e/2008-08-26_145659n.jpg

After next week, I'm planning various tests.

Ganryu

 

[Dennis Wood]

Great. We've still got the SS4200 here, but it's on the shelf with no drives installed.

I just checked our calculations spreadsheet and we were able to get 51MB/s write and 72MB/s read from the SS4200, but this was only in RAID 10 mode with 4x 1 TB 7200 rpm drives. This was only possible with a test workstation running Vista SP1 (SMB2) with a 3 drive RAID 0 array. This actually was a promising option for us, but the fact that 4x1TB drives only provide a 2TB volume was hard to take. In contrast, the Qnap unit provides a 4TB drive array in RAID5 with write and read speeds similar at 45MB/s and 60MB/s. In other words, the Intel box is the fastest fault tolerant array we've tested, but suffers in terms of lost capacity if you use RAID10.

 

[ganryu-island]

Thanks, Dennis,

I understand your test environment as follows.
TS-509 : default 1GB cache
SS4200-E : default 512MB cache
I think you improve a performance when cache of SS4200-E is set to 1GB.
Let me know your test environments(CPU_clocks, memory_size, NIC).

The following is the test result(by CrystalDiskMark) of Synology_DS508, for your reference.
http://www.ganryu.org/DS_pf_Check/ds508pf0.htm
I don't have TS-509, but could you do a benchmark test of TS-509 using CrystalDiskMark ?
http://crystalmark.info/software/CrystalDiskMark/index-e.html

Ganryu

 

[Dennis Wood]

Upgrading the cache would improve performance for files under 1GB in size...but after it's used up, the unit must be able to maintain a given physical write/read speed. For our applications, the typical file is typically 1 to 200GB..so all we're interested in is non-cached performance.

The workstation: Asus P5W-DH motherboard with onboard PCIe Marvel dual NICs, 2 GB RAM, Core 2 Duo E6400. The workstaion has a RAID 0 array using 3x320GB WD drives on ICH7 (onboard controller).

The values from the CrystalDiskMark are relatively fine for comparison, but they are much higher than what you'll actually see in measured performance with file sets over 2GB. The write rate of the unit is about 50MB/s max with larger files.

 

[ganryu-island]

Thanks, Dennis,

I found out that a performance of TS-509 is very good.
Various tests of SS4200-E are written here after next week.

Ganryu

 

[Gidi]

Max mem/cache support

Just had a look at the linux startup of the ss42k with the current version
of the firmware (1.1.11.32736) and saw this

Warning only 896MB will be used.
Use a HIGHMEM enabled kernel.
896MB LOWMEM available.

As I wrote before I had installed a 2GB DIMM in the box and could not see
any real performance improvement.

Looks like we need to replace the kernel

 

[Gidi]

SS4200e and ZFS

Just run iozone on my SS4200e loaded with OpenSolaris 2008.11, 2GB Memory, 4 Samsung 1TB F1 drives as a raidz volume.

The iozone parameters where the same (from Tim) as a few posts before.

I had already about 600GB data in that raidz pool and hope this doesnt change the result to much.

The configuration
Vista Ultimate 32bit SP1
HP/Compaq DC7800 (Q9300/2.5GHz/4GB Ram)
Intel gigabit ethernet chip
Procurve 1800-8G Gigabit Switch
SS4200e (OpenSolaris 2008.11/2GB Ram)

Vista gives me never a constant network transfer rate to any of the other
boxes I have (mostly notebooks or some old crap). Eventually I'll rerun the
test on a XP box some times