Poor FC performance in comparing with competitor

[Warlib]

Hello,

I've tested Open-E 6 Lite on the 2x Xeon Supermicro Server with LSI 9625-8i raid controller (http://www.lsi.com/products/storagec...AS9265-8i.aspx) with 8 pcs 600 Gb SAS 15K HDD and QLE2562 ISP2532-based 8Gb Fibre Channel to PCI Express HBA. IOMeter result is about 80 MB/s (read). Our existing IBM DS 4700 shows about 40 MB/s (read). Test was conducted from the same IBM HS22 XM server with Brocade FC switch.

But, one of the competitor http://www.avroraid.com/en/technology/testing-results claim that they have got about 917 MB/sec at the weaker (cheaper) server configuration. What's wrong with Open-E? Why performance is so poor? What can I do to impove performances?

Thank you?

[adamvh]

iometer has alot of knobs and dials.

probably find your iometer test settings are not identical, there are alot of setting in iometer than can skew the results alot!
reading marketing material is great, but you need to use the identical iometer setup as them if your trying to compare.

[ellaharris]

This is an interesting topic. IBM is first to deliver an FCoE-ready 10GbE blade switch for converged data and storage networks inside the chassis for maximum consolidation.

[Al-S]

The Lite version has some limitation, as it does not support 10 Gb Ethernet, or Multiple CPU Support, etc.. for more details about different between Free Lite version of DSS and the full DSSv6, please visit this link: http://www.open-e.com/products/open-e-dss/comparison/

Also you said you have 8x600GB HDD, that means you have 4800GB (4.8TB) ? as the Lite is only designed to deal with 2TB of storage. while the Full DSS can deal with unlimited size.
So please try with the full DSS version, also try to tune your system, to match the same as other products, and please try to test both, as all can claim, right!

[Warlib]

We installed trial version of DSS 6.0 and get additional information concerning IOMeter options from AVRORAID. They got test using file pattern with block size 1 Mb (100%). We used the same pattern and repeat test. The result is better, but still is worse than AVRORAID 280 MB/s. Specialists from AVRORAID believe, that Open-E can't work properly (with high performance) with large block size (1MB).

[SeanLeyne]

...Specialists from AVRORAID believe, that Open-E can't work properly (with high performance) with large block size (1MB).

It is not surprising for a vendor to try to dismiss another vendor's products.

One thing to consider... does the AvroRAID product support the same feature set as Open-e? Volume Snapshots?

Don't get me wrong, I think that AvroRAID has some potential (I had a look for myself, too bad they don't have a North American base/var network). I just want to make sure that you realize what the trade-offs (every solution has them) are between the 2 solutions.

[SeanLeyne]

...But, one of the competitor http://www.avroraid.com/en/technology/testing-results claim that they have got about 917 MB/sec at the weaker (cheaper) server configuration....

I just re-reviewed the performance number from the site and I can't find any matching details for the config which provided the 917MB/s numbers.

In general, I have to say that unless you run the performance test yourself using a single set of hardware, trying to compare your results with theirs is not advisable. There are all sorts of reasons why the numbers could be different.

I would contact them and see if you can get a trial/eval version so you can run the tests yourself.

[Warlib]

Of course I understand, that competitors always try to persuade, that their decision is much better. :-) And of course I understand, that features of AVRORAID is poor, than Open-E. But I would like to know what's wrong with Open-E performances. Is it normal or not. Probably something wrong with hardware tuning or something else. Unfortunately, I couldn't find some good test results of Open-E. And unfortunately, it's rather hard to make head-to-head comparision, since AVRORAID support only ATTO target FC controller, but I have only QLogic. Probably someone knows where to get Open-E performances test results?

[Al-S]

Try to use the full build of DSSv6 ( not the LITE, as you have a storage size more than 2TB ).
Also update your RAID Controller Firmware to the latest build (check your controller website for that).

[To-M]

Ah the world of the competitive market , looking at AVRORAID's site, I dont even see a Forum or a Blog... and when I look at there Hardware Compatibility List it is very limited where we have over 850 products that we support. I first tried to go to there support and looks limited - take a look at this link and compare. Keep in mind we are a unified file and block storage management operating system so we will have more feature sets and RAID sets as well.
http://www.open-e.com/service-and-support/

I would recommend to use the Full version of the DSS V6 and make sure the RAID controller has the Write Back settings enabled, also the links below.

Read this concerning the tests we did w/ a FC Volume created on DSS V6 with FC HBA dual 4Gb and MPIO and the results where 772 MB/sec, so w/ one FC port it will be less due to the MPIO but still you should be getting better results - follow this article that will show how to setup the IOmeter.
http://blog.open-e.com/a-few-practic...about-iometer/

This link below might help explain the "random-vs-sequential" w/ IOmeter.
http://blog.open-e.com/random-vs-sequential-explained/

Learn the Write Cache settings for your controller.
http://blog.open-e.com/just-how-impo...s-write-cache/

If you are testing the full version send in a request at: Pre-sales (at) Open-E (dot) com and we will look at the logs but please look at the links first.