DSS / iSCSI Performance

[markstaller]

This is good stuff and I agree we should be posting more speed and setting results. The only problem is we deal with so many resellers and customers and the results are never the same.

It's quite clear that these results differ and aren't comparable, but I guess it stills gives some insights on what can be expected for a certain combination and it's good at least for building some rules of thumb.

I also wanted to let you know that the DSS requires CPU 3.0GHz or faster (higher cache the better) yours is 2.66GHz.

Really, I didn't thought that a CPU above some P4 on a DSS would have a significant impact on iSCSI-I/O (significant compared to Disks, Raid). You mean the bigger 2nd-Level cache on the CPU, not RAM right?
Anyway, also for this some benchmark for comparison would help to decide for anybody what's worth it..

As with any new products and software you always want to provide latest and fastest as the product is designed for this.

Sure, this box is ~1y old (this is also not the box that will have high IO in future) but it's my test-environment for now..

Also try to set your RAID to highest chunk size 256+.

-> just changed the stripe-size from 64k to 256k; will test afterwards.

Check the CPU, MEM and NIC stats under Status - Hardware - Function: Server statistics to see what’s happening.

This returns an empty page on my DSS ? I just didn't had the time to complain so far

By the way I had many engineers call me and tell me their is a firmware update from
3Ware to fix some performance issues - this is to be released in Q2 of 07.
Another note, allot of engineers had reported that the Areca 1280 ROCKS on performance even on a RAID 6 is only 5% less then the fastest 3Ware RAID5!!!

I also did some research concerning the 3Ware and they have some speeds with charts for RAID5 and others with Databases and Webservers.

http://www.xbitlabs.com/articles/sto...-9500s8_8.html

Although I guess I'm not after a controller problem when suffering from 40MB/s read in a 7-disk RAID5 while writing 60+, it's interesting. It simply tells me to buy another controller that isn't broken under this or that circumstance for nearly the same price..

Regarding NIC and iSCSI-Parameters honestly I'd prefer to rely on whatever your engineers found to be the best (which isn't true for the initiator side for sure..). All of such fine-tuning is extremely time-consuming and leads to complex settings and mostly unreproducible results when adding i.e. another initiator.. I'm happy for now when reaching 80% of whats possible with 20% effort - if you know..

Also keep in mind that the reason other vendors don't like to report to much on speeds is the variables involved - unless you pay for their proprietary hardware and software $$. We are a fraction of the cost of standard equipment sets like EqualLogic (starting in the $40k+++) which by the way does not offer you the choice that we supply for all the different compatibility listings for DSS - which is extremely inexpensive.

That's why I'm with DSS I can solve any performance issue with buying a 400% oversized system always, but the glue for my "just in time" datacenter is to have what I need now and nothing more, that saves thousands each year.
Also on another note, not all of these "enterprise" systems are that drop fast.. I did perf-test with several E**, N***** boxes before and most of them underperformed by ages as well (but for sure for another 30k$ you could buy FC-disks and x more controller to compensate that..)

I'll try to get some more results but resources are - as always - limited.. For the next two DSS planned I'd wanted to have a better understanding where to invest; not buying the most expensive from everything, which also helps but costs much more money than focusing on the relevant parts.

Michael