Intel SS4200-E speed

[dkazaz]

Hi All

I just bought an Intel ss4200-E with 4 Seagate Barracuda 1TB drives (normal, not ES) in RAID 5.

So far am pretty pleased with it in terms of ease of use, convenience and functionality. The latest firmware seems to address many of the issues mentioned in the recent review on this site.

However, one area where I'm slightly disappointed is transfer speed. On a gigabit network I get about 25-30MB/Sec, 31MB max.

I was hoping for the 48MB/sec listed in the NAS charts. My files are large mostly between 15MB - 1GB.

I wonder if there is something affecting it?

• I have turned on write caching (anyway its connected to a UPS so it does so automatically)
• I connected the PC to NAS via a 3com 5-port gigabit switch (the plain unmanaged model). The switch supports jumbo frames but of course the NAS doesnt.
• I'm using Cat 5E cables and windows shows the connection as Gigabit.
• My only thought is that my OS is Vista x64 SP1. I know Vista has issues with file transfers and I thought SP1 solved those issues but if not, I may have to re-install XP (I'd rather not since I just got used to Vista's weirdness) I have an XP HTPC, maybe I should test that?

Frankly I can't think of anything else. Can any of you wise folks suggest anything?

I'd really appreciate any advice. Thanks!

 

[thiggins]

How are you measuring the speed? Our testing is done with iozone, which has very low overhead. Are you measuring read or write?

If you are measuring via file copying, your measured speed may be lower. File sizes also make a difference.

If you use the NAS Chart and plot the throughput vs. filesize, you can see that larger filesizes have throughput down in the 30 MB/s range.

 

[dkazaz]

Hi Tim, thanks for your reply!

To measure speed, I just looked at the windows vista file copy dialog for transfer speed (during writes).

I didn't think to use any other software. I will look into using iozone to test my system.

My files fit squarely within the category you call 'large files' where the charts indicate a 40-48MB/sec transfer speedin your tests. However the larger files (300MB-1GB videos) were copied at about 20-23MB/sec, sos till comparatively slow I think.

I will use iozone next, if I can get it to run on vista 64 and report back. I assume you don't see much merit in my notion that the switch or OS were at fault?

 

[thiggins]

To measure speed, I just looked at the windows vista file copy dialog for transfer speed (during writes).

First, sanity check that number by timing the transfer and dividing into the file size.

I will use iozone next, if I can get it to run on vista 64 and report back. I assume you don't see much merit in my notion that the switch or OS were at fault?

Doubtful. I think it may be more a case of comparing apples and oranges (the different test methods.

What do you get if you just copy from one computer to another?

 

[dkazaz]

I have at least verified that this is not down to the OS. I tried several similar tests from XP and again got about 28-29MB/Sec max on the same file-sizes.

Nest I will try iozone but I'm a little intimidated byt the all the command line options- I need time to learn the usage first.

I don't suppose there a simple command I could just type into a cmd prompt?
I have little patience with reading documentation these days - its too hot!

 

[thiggins]

1) Install iozone.
2) Create a mapped drive for a folder on the 4200.
3) Open a command window and enter this command line:
iozone -Rab [results file name] -i 0 -i 1 -+u -f [path to directory on NAS device under test] -q 64k -n 32M -g 1G -z

For example:
iozone -Rab intel_ns4200e_1g_4kjum_r5_1_00_01.wks -i 0 -i 1 -+u -f i:\temp.tmp -q 64k -n 4k -g 1G -z

 

[dkazaz]

Thanks for the guidance, I really appreciate it.
I finally got around to running the test (sorry for the delay but work sometimes takes over)
I uploaded the results in the zip in case you're interested.

I'm trying to interpret the results now, but the numbers seem strange if I'm reading them right - really wide variation. If these are MBs then it goes from about 30MB/s sec to 800MB/sec?! Or should I divide by 10?

And what is the record size? If my understanding is correct, it refers to the size of data packets the application or OS (not sure which) reads and writes.
Is it something that can be controlled since it seems to have a massive impact on performance.

Any input would be appreciated. Thanks again!

 

[dkazaz]

OK, I did some more research into iozone to learn how to interpret the results file. I assume the NAS charts published use the 8K record size since that is (a) used by windows for network transfers and (b) mostly in line with my results.

Now for the eird part:

1) My results are a bit higher than the review on the write tests (could be due to my system spec being higher than the test PC - intel quad with 4GB RAM). They were also much higher than windows reported rates.

2) I have to say though, that the windows numbers seem relatively accurate based on rough calculations of file sizes vs. copy times.
Could mean that iozone somehow is more efficient at copying files than windows? In principle it wouldn't surprise me but I would have thought it uses the same low level functions of the OS to do the job?

3) The picture is very different on the READ side, where my results are a solid 10MB/sec lower than the test on every file size. This was especially surprising as reads are supposed to be less demanding. I would really like to figure out if there's a way to fix this.

4) A key difference between the charts and my test is that I'm using newer firmware - the new version corrects some serious issues but could have an impact on performance.

All in all, I'm still ahppy with the performance but I'd love to find a way to fix those strange performance issues. I've even thought about upgrading the RAM on the NAS to 1GB but I don't exactly what RAM it uses (is it ECC?) so I'm a bit weary of pinning my hopes on this...

 

[BostonDan]

Memory?

I don't know if I'm supposed to post a new thread of keep all threads relating to speed within this thread, so apologizes if I'm posting this in the wrong place. Has anyone upgraded the memory of the SS4200 and if yes, what size memory did you upgrade to and is there a big difference in performance? Was it worth the effort & time?

 

[dkazaz]

I think you better point this to intel support or give it a try yourself since memory cost is in the region of $25 only.

 

[thiggins]

I'm trying to interpret the results now, but the numbers seem strange if I'm reading them right - really wide variation. If these are MBs then it goes from about 30MB/s sec to 800MB/sec?! Or should I divide by 10?

Thanks for uploading the zip of your results. iozone reports filesizes in Bytes and throughput in KBytes/s.

And what is the record size? If my understanding is correct, it refers to the size of data packets the application or OS (not sure which) reads and writes.
Is it something that can be controlled since it seems to have a massive impact on performance.

Record size is controlled by the filesystem. Some OSes may let you futz with it.

 

[thiggins]

1) My results are a bit higher than the review on the write tests (could be due to my system spec being higher than the test PC - intel quad with 4GB RAM). They were also much higher than windows reported rates.

This would definitely affect performance. Iozone results are closely tied to the system and OS that it runs on. That's why all NAS Chart tests are run with the same system...even though it's getting a bit long in the tooth at this point. I would expect your system to produce higher results for write, and it does, both due to the higher memory and faster processor.

I plotted your results and the NAS Chart results using the 64 KB record size and have attached the write and read plots (thumbnails below). You can see cached performance (> 125000 KBytes/s) up to the 256 MB filesize. But then your results drop to meet mine at 512 MB and 1 GB, which are showing hardware-limited (non-cached) performance.

2) I have to say though, that the windows numbers seem relatively accurate based on rough calculations of file sizes vs. copy times.
Could mean that iozone somehow is more efficient at copying files than windows? In principle it wouldn't surprise me but I would have thought it uses the same low level functions of the OS to do the job?

iozone is very efficient and will tend to provide a higher results than OS filecopies, at least for SMB.

3) The picture is very different on the READ side, where my results are a solid 10MB/sec lower than the test on every file size. This was especially surprising as reads are supposed to be less demanding. I would really like to figure out if there's a way to fix this.

Looking at the read comparison plot, our results track pretty well up to 256 MB filesize. Your results do drop after that and I don't know why. Most likely OS effects.

4) A key difference between the charts and my test is that I'm using newer firmware - the new version corrects some serious issues but could have an impact on performance.

Yep. It could. But, again, when you make a proper comparison, our results are not that different.

All in all, I'm still ahppy with the performance but I'd love to find a way to fix those strange performance issues. I've even thought about upgrading the RAM on the NAS to 1GB but I don't exactly what RAM it uses (is it ECC?) so I'm a bit weary of pinning my hopes on this...

I don't think there is anything to be "fixed". Adding more memory might improve write performance under some conditions. But I'd be happy with what you have and get on with just using it!

 

[dkazaz]

Thanks for all your advice Tim!

You've made everything much clearer!

As for 'fixing'performance - agreed. The product works as advertised - a rare pleasure in today's tech industry! No reason to complain.

When I have time I will run iozone from linux (once I've installed it on this PC). In the long run the only performance that matters is the one under the OS one uses most of the time.

I may still go for the 1GB RAM - DDR 2 prices are so low these days. Do you know if the box is using standard PC4200 or something special e.g. ECC or SO-DIMM? I'm going to avoid opening it up for a while - I'm enjoying using it too much right now!

 

[johnno]

Thanks for all your advice Tim!

You've made everything much clearer!

As for 'fixing'performance - agreed. The product works as advertised - a rare pleasure in today's tech industry! No reason to complain.

When I have time I will run iozone from linux (once I've installed it on this PC). In the long run the only performance that matters is the one under the OS one uses most of the time.

I may still go for the 1GB RAM - DDR 2 prices are so low these days. Do you know if the box is using standard PC4200 or something special e.g. ECC or SO-DIMM? I'm going to avoid opening it up for a while - I'm enjoying using it too much right now!

It's just normal DDR2 SDRAM. I've installed a 1GB stick of Kingston 553mhz DDR2 ValueRam (CL4) without issue (i've not used the intel firmware as of yet, so I can't say whether it helps performance or not).

 

[dkazaz]

Great, thanks for clearing that up - I will do the same! Did you see much performance impact?

Regarding the firmware, it is advisable to install it as it resolves some serious issues - have a look at the intel support pages for this product to see if it makes sense for you.

 

[johnno]

Great, thanks for clearing that up - I will do the same! Did you see much performance impact?

Regarding the firmware, it is advisable to install it as it resolves some serious issues - have a look at the intel support pages for this product to see if it makes sense for you.

The 'firmware' as Intel likes to call it is just a custom linux distro contained on a 256MB disk-on-module connected on the IDE port.
If you're using another OS (WHS for example), it has no effect.
(now if they wanted to put out a BIOS update, I'd be mighty interested -- too many unpleasant bugs in the current one if you ask me).

 

[dkazaz]

The 'firmware' as Intel likes to call it is just a custom linux distro contained on a 256MB disk-on-module connected on the IDE port.
If you're using another OS (WHS for example), it has no effect.

That makes sense. To be clear, are you talking about using a different OS on the device? Did you disconnect the DOM and connect another drive? And if so running WHS? I was under the impression MS has specifically advised against using it with RAID arrays.

(now if they wanted to put out a BIOS update, I'd be mighty interested -- too many unpleasant bugs in the current one if you ask me).

I'd be very interested to hear what they are - I have found no info on the net whatsoever on this topic, can you please elaborate? Or point me to a link?

 

[dkazaz]

OK, I found the reseller threads where the issues of customization are discussed, but for the intel packaged model I think only some apply. There are issues with customization and installing WHS etc.

However I saw no issues relating to data loss which would be the key concern. Am I wrong? So far all I've perosnally found is some weird behaviour with regards to connecting with clients and since I'm on Vista 64 which is not supported yet, I can't be sure these are not due to this...

 

[Gidi]

I've bought also one of this boxes together with 4*1TB Samsung F1 drives.
Connected to a Netgear 8P Gigabit switch and a AMD 4600 X2 Box XP-SP3 with a Broadcom GB Ethernet.

Use them as a RAID5 set and was suprised by the fact that during the first
reconstruction the box was fully usable and I could start to load it with tons of files (500-1G each).
After about 6.5Hr the process was completed

Copy/robocopy/rsync performance was also a bit of a mixed bag.....between excellent and ok (25-75MB/s). DD shows a read/write performance of up to 85MB/s with large files. I've also upgraded the memory with a Corsair 2GB-667 DIMM and could not measure any performance change.

I'm currently looking for a way to have OpenSolaris or Nexenta (ZFS) on that box (have a 8GB DOM ready ) and test a setup on a Intel Box with 2 160GB Samsungs with horrible performance compared to the SS42K.

Regards
Guido

 

[dkazaz]

Interesting that your memory upgrade did have any benefits, though your results were pretty impressive to begin with. I read that the RAM can be upgraded to 1GB, so perhaps this may have something to do with it?
Personally I can't be bothered especially if the benefit is so limited. Of course in your case, it makes more sense since you plan to run another OS.

I'm really surprised by the 85MB/s on large files - my tests showed performance dropping in large files. I will try to run iozone on a linux and an XP box to see how much difference it makes from Vista.