Sign in with
Sign up | Sign in
Your question

PCI Express Battles PCI-X

Last response: in Storage
Share
November 24, 2005 11:15:16 AM

I'll get everyone off side by saying "What a half ass'd review."

Whats with the 'crashed'. Please explain the behaviour, conditions and results. If it's unreliable, it fails. At the very least make sure there is no tail on the graphs where it failed.
Need a summary of PCI-E vs PCI-X results or somewhere to visualise trends.
Need more of a variety of tasks to test different aspects.
Avg. response times (random read/write/read+write).
sequential read (short/buffered, long/disk limited).
sequential write (short/long).
sequential read/write with mixtures.
You say about multiple cards in the one system would show a difference (and >8 drives) but I don't see any results ie. test your hypotheticals to make them fact instead of guessing.
Test card IO performance by reading data that fit's into their cache then perform lots of reads from what is in the cache (OS cache off).
System benchmark that is IO sensitive that may reveal differences in efficiencies between card&driver&MB combo's. eg. CPU utilisation ?
Graphs are too cluttered. Add to these … Pick the best of each and compare. Pick two cards that are very similar except for interface and compare.
Can you give a date of release or some other indication apart from firmware/driver revision numbers as to their age/maturity.
The comparison table doesn't properly indicate which use hardware RAID5 XOR. I'm sure the Broadcom does and you say none. At east say integrated or repeat the name of the chipset or main identifier. Some my use one chip, others use an IO processor and an interface chip, others may require 3. eg. The Intel IOP331 has XOR and IO processor in single package which isn't clear in your description.
Get more help from tech support from the companies as to the most accurate/complete way to describe their products for the spec sheet (eg. comparison table). Berate them in public if they don't help.

BTW: I skimmed through the first bit and went straight to the last three pages. It's midnight here so I'm off to bed to regret my comments later. In the meantime, discuss with pleasure.

Sent to webmaster
Quote:

Need to create a thread in the forums per news/article and link directly to it. At the moment:
* very few people see the comments regarding an article;
* less likely to leave a comment because they have to find the thread or start a new one;
* several threads per article because their didn't see that it was already there.
November 24, 2005 3:24:04 PM

I agree, there is a lot missing from this review. The main advantage of PCIe is the dedicated bandwidth of each device and this was not tested. The test should have compared multple devices being used by pitting PCI-X RAID with a PCI-X NIC against a PCIe RAID with a PCIe NIC combo.

If you want to give a storage system a workout, try loading and editing uncompressed 1080p HD video. You'd need a high end capture card with the storage systems where you need to have 2Gb/s sustained bidirectional support.
November 30, 2005 2:36:37 AM

This review would make the Bush Administration proud. Did anyone actually look at the graphs? (I know, lots of miniscule lines, as per the comment above) There isn't a scintilla of difference between the performance of the 2 RAID cards in question. In THEORY, PCI-Express would be better with it's serial transmission method; but the graphs don't support this.

>>Heckuva job, Tommy!
!