You actually used 2GB, at least in the 4KB RR test. The file shows...Originally Posted by SteveRo
You actually used 2GB, at least in the 4KB RR test. The file shows...
dammit! - check some of the othere for me.
I created the full disk file but the runs didn't use it?
Does - Maximum Size 4000000 mean 2GB?
So 6GB would be 3x - 12000000?
12GB - 24000000?
OK so new iometer files - just two as a sample - Areca 1880/4GB 16xAcard 9010 with the runs using 48000000 sectors or 24GB - performance w/o the cache does not appear so good -
https://docs.google.com/leaf?id=0B2A...thkey=COXR1d4O
what was the 4krr qd1 with the larger testfile? i see a database test there....but was there a straight 4krr qd1 on the 1880?
"Lurking" Since 1977
Jesus Saves, God Backs-Up *I come to the news section to ban people, not read complaints.*-[XC]GomelerDon't believe Squish, his hardware does control him!
@CT
There was an issue running the random read tests on both the PCH and the Areca, don't know why, there really shouldn't be any issues with that config.
QD1 (database) using the large testfile was close to 100MB/s, not bad at all.
The Areca results are not so impressive outside cache, compared to the previous runs using the PCH that is
The latest runs are fine, you don't need to run 100% span, 16GB is sufficient and that might improve the results a few %.
database_qd128_sro_acard_areca.png
fileserver_qd16_acard_areca.png
I've performed a few tests on the 9265 and the results are coming up within 30min or so.
-
Hardware:
Anything with cache is going to be slower with ACARDS. ACARDS are so fast that adding another layer data must travel to just slows everything down. Iodrive manual mentions never to use caching with it since it would just be slower.
Meaning that when X79 enters, Steve's ACards will be really extreme
The 9265 is pretty extreme, not so much at QD1 though.
w/o FastPath, 8 Intels resulted in
235201 4K iops
161972 8K iops
and almost 100K iops at 16K
rr_9265_8r0_Intel.png rr_9265_8r0_Intel_iops.png
Last edited by Anvil; 04-17-2011 at 04:45 PM.
-
Hardware:
At higher qdepth the 9260 with or w/o fastpath looks to be a huge performer.
What is the highest typical workstation qdepth, seems like I saw a number once - 5 or 6?
Has anyone run himon while executing pcmv or pcm05?
Curious what the qdepth looks like for those.
Yes, w/o the FP key 9265 needs high QDs to perform but thats what its made for, a large number of users.
I'm waiting for the FP key and that should make a difference, Computurd should have run some of these tests if he's still got the SSDs and the controller
If you look in the hIOmon thread I'm pretty sure I did monitor a PCMV HDD test, all info should be in the screen-shot.
I'll see if I can find it later today, from what I recall it had a pretty high QD. (could be a short burst)
Comparing the ioDrive vs the 9265 with 4R0 Vertex 2 and 8R0 Intels
rr_ioDrive_VS_9265_woFP.png
Last edited by Anvil; 04-18-2011 at 02:37 AM.
-
Hardware:
Steve
About the PCMV monitoring, look at page 7 in the hIOmon thread, starting at post# 173 and going on into the next page
-
Hardware:
Hi Steve,
In post #174 Anvil came out with an overall average read I/O operation QD of 2.033 and a maximum of 73. For writes it was QD 1 for avg and max. This was recorded at the device level.
In post # 178 Anvil came out with the following for the HDD suite test. (Again at the device level). read avg 2.572. Max 77. Write avg 1.169 and max = 13.
EDIT:
I monitored xfer sizes with PCMark at the device level. You can see the output file here: (It is too large to view online, but it can be downloaded).
http://cid-541c442f789bf84e.office.l...e.aspx/.Public
Last edited by Ao1; 04-18-2011 at 07:58 AM.
^^ Thanks guys, I will read up!
i still have the drives and the keys. i will run these shortly
"Lurking" Since 1977
Jesus Saves, God Backs-Up *I come to the news section to ban people, not read complaints.*-[XC]GomelerDon't believe Squish, his hardware does control him!
Thats great CT!
You'll find the profiles at this link
-
Hardware:
running all of them now anvil. pm your email addy and i will jsut email the results, easier that way
"Lurking" Since 1977
Jesus Saves, God Backs-Up *I come to the news section to ban people, not read complaints.*-[XC]GomelerDon't believe Squish, his hardware does control him!
http://cgi.ebay.ca/Fusion-io-ioDrive...item3a656ec3f7
damn this is a good price. $2650 for a 320gb iodrive. Those are 10k new. It is used however and does not say if its MLC or SLC. Looks SLC. Looks stolen. Perhaps I will drive down to check it out since its 5h away from me... Too bad the capacity would be lost on me since I would just use 80GB of it to raid it with my iodrive haha.
Last edited by One_Hertz; 04-28-2011 at 04:17 PM.
who cares if its stolen. buy it for the betterment of mankind
"Lurking" Since 1977
Jesus Saves, God Backs-Up *I come to the news section to ban people, not read complaints.*-[XC]GomelerDon't believe Squish, his hardware does control him!
5 hour drive, might be worth the look
Is there some way to know how much "life" there is left on the NAND?
Isn't the 320GB faster than the 80GB?
-
Hardware:
The device firmware records how much data was read/written and the driver displays that info. These things basically last forever though due to grade 3 enterprise SLC NAND and their wear leveling... Something like 40 Petabytes of writes for the 80GB version, so 160PB for the 320GB unit. Reliability is one of the major selling points.
I do have $3k to blow on useless stuff right now, but an IOdrive that large.... I don't know.
Yes, the 320GB is faster at writing than the 80GB by about 200MB/s.
If you flash the firmware the read/writes are set back to zero.
I bet One Hurtz is in his car right now.
I'm also betting that Intel used IOmeter for this bench, so I guess this belongs here. Interesting anyhow.
nice graphic. i am of the thinking that when NCQ is in full effect the drives accesses are 100 percent random. this stacking of commands and paralleling of them would result in random I/o. however, random I/O over several channels, as we see, can be blistering fast in regards to small file I/O.
"Lurking" Since 1977
Jesus Saves, God Backs-Up *I come to the news section to ban people, not read complaints.*-[XC]GomelerDon't believe Squish, his hardware does control him!
WOW he dropped the price down to 2k. I'll get in touch with him on Monday and buy it as long as it is SLC
Last edited by One_Hertz; 04-30-2011 at 10:58 AM.
Where are the results?running all of them now anvil. pm your email addy and i will jsut email the results, easier that way
Reply With Quote
Bookmarks