View Single Post
Old 10-22-06, 11:05 PM   #4
fhj52
Registered User
 
Join Date: Jan 2005
Posts: 135
Default Re: What is the NVRAID stripe size?

So this is what I am talking about:
Code:
iozone -a -i0 -i1 -s 8G -r 128 -r 256 -r512 -Rb /tmp/iozone-NVRAID_JFS=8Gfile_test.xls

        Run began: Sun Oct 22 18:46:49 2006

              KB  reclen   write rewrite    read    reread    
         8388608     128  134176  142128    27766    95556
         8388608     256  138216  143879    28421   105634
         8388608     512  136698  133830    29007   110574

iozone test complete.
Prior tests all show the same thing: for 8GB file, the bottom drops out of the READ as soon as record size hits 128kB. The reread has never been more than 10% higher and, typically, for large files is about the same as the READ since there is not much cache to use.
Smaller record sizes READ at > 112MB/s for the 8GB file. Also, smaller files don't have the dropoff AND it appears to be only for the READ.
Here's a snippet of the data from the full test showing just the READ:
( The top row is records sizes, the left column is file sizes )
Code:
	4	8	16	32	64	128	256	512	1024	2048	4096	8192	16384
1048576	1795491	1857415	2328637	2141122	2387256	2429762	1690177	1383961	988808	1111140	1027648	1115923	911462
2097152	1960697	1960180	1885432	1864134	2303706	2265836	2247662	1547866	1195035	1190274	1142225	1126850	1116831
4194304	135713	215898	201045	193421	165662	148444	188800	96383	70322	117336	119855	155411	132248
8388608	115405	115447	115406	115429	115366	26721	27894	28764	28264	28563	28467	28552	28344

Same test on a SCSI disk(no RAID) shows no big hit:
Code:
iozone -a -i0 -i1 -s 8G -r 128 -r 256 -r512 -Rb /tmp/iozone-SCSI_15k_ext3=8Gfile_test.xls

        Run began: Sun Oct 22 14:15:33 2006

              KB  reclen   write rewrite    read    reread
         8388608     128   62827   52685    46054    59147
         8388608     256   64989   62218    49573    58292
         8388608     512   64537   49897    52131    57109

iozone test complete.
Smaller and larger records all maintained about the same I/O.

Although, oddly, WRITE still outruns READ on all the tests, the results are pointing directly at NVRAID as the cause of the problem.

It appears to be NVRAID controller but there are other things that could be the cause. Tests on the ATA drive are running ...

Edit: IT APPEARS to have the SAME problem as the 128kB record READ is 23MB/s. That's a clear indication to me that the controller is the problem with, of course, possible contributing factors.

ATA drive exhibits the same problem, although the difference between normal and the drop is not as drastic.
Code:
        Run began: Sun Oct 22 21:51:28 2006

              KB  reclen   write rewrite    read    reread    
         8388608     128   44676   36530    24279    40410
         8388608     256   43235   36132    26239    41244
         8388608     512   44912   33881    27399    38586
I need input to determine what causes this.


And is there anybody out there running NVRAID that can verify the same anomaly? Who knows, maybe it is the controller on this mobo ...

Thanks!
__________________
When two people meet and exchange gifts, each has one object.
When two people meet and exchange ideas, each has two ideas.
... Open Source. Just do it.

---------------------------------
System: BFG GTX260^2 graphics but has ** TERRIBLE BLINKING OS **
SuperMicro H8DCi+AMI BIOS;dual Opt'285;8GB;LSI 320-2x w/ 6xU320 Fuji' MAXs in RAID 10; 4xSATAII on LSI 3041E for backup. Multi-boot Mandriva Linux, openSUSE, WinXPx64 & Win2k-AS; Creative Audigy2-Digital audio.
Gigabyte GA-2CEWH & NVRAID are GONE ... Finally!!

Last edited by fhj52; 10-23-06 at 12:06 AM.
fhj52 is offline   Reply With Quote