APLawrence.com -  Resources for Unix and Linux Systems, Bloggers and the self-employed

amird driver, BDFLUSH and NAUTOUP, and more

I recently installed a new SCO 5.0.7 server to replace an older SCO 5.0.5 system. We put in new hardware, with an LSI Ultra-320 raid controller. Two drives were configured for hardware mirroring. That hardware uses the AMIRD driver, and the installation was quick and simple.

I checked disk performance against the old server with various permutations of "timex dd if=/dev/root of=/tmp/this bs=1024 count=8192", using different block sizes and counts, and also adding some seeks. I did these on both machines, with some tests including intermediate "ls -lR /u" to clear the buffer cache and some not, and in general I could say that the new machine was 6 times faster for disk access and a little more so for buffer cache accesses.

Imagine my surprise when the customer reported that it seemed slower to them.

Well, I knew that /etc/idrc.d/amird might be at fault, so I added an "exit 0" at the top of that script and killed the /etc/amirdmon process that it had started. That helped a little, but they still complained of sluggishness. I could see that in "sar -d", too:


12:00:00 Sdsk-0    26.95      1.00      0.67      5.30      0.00    400.15

12:20:00 Sdsk-0     9.58      1.00      0.82      2.92      0.00    116.44

12:40:01 Sdsk-0     9.65      1.00      0.72      1.92      0.00    134.11

13:00:03 Sdsk-0     5.81      1.00      1.17      5.15      0.00     49.57

13:20:00 Sdsk-0     6.28      1.00      0.42      1.07      0.00    148.58

13:40:00 Sdsk-0     2.10      1.00      0.12      0.27      0.00    179.93

14:00:00 Sdsk-0    24.20      1.01      1.72      6.52      1.44    141.05

14:20:00 Sdsk-0    29.65      1.01      0.57      1.42      5.85    517.83

14:40:00 Sdsk-0    29.33      1.00      2.12      6.13      0.00    138.62

15:00:00 Sdsk-0    15.44      1.00      1.16      2.98      0.00    133.49

15:20:00 Sdsk-0   100.00      1.01      5.61     19.16      2.28    322.97


Average  Sdsk-0    48.04      1.02      1.43      6.29      6.55    336.06
 

I couldn't get the command line scodb to work for me over my ssh connection, so I had to use /etc/conf/cf.d/configure to modify BDFLUSH and NAUTOUP to "1". . Relink, reboot, and sar was much happier:

15:38:41  device   %busy     avque     r+w/s    blks/s    avwait    avserv (-d)
15:38:56 Sdsk-0     0.20      1.00      1.93      4.40      0.00      1.03

15:39:11 Sdsk-0     0.07      1.00      2.13      7.73      0.00      0.31

15:39:26 Sdsk-0     1.53      1.00      3.40     10.93      0.00      4.51

15:39:41 Sdsk-0     1.33      1.00      3.40     11.59      0.00      3.92

15:39:56 Sdsk-0     1.00      1.00      1.33      2.93      0.00      7.50

15:40:11 Sdsk-0     1.60      1.00      4.27     13.87      0.00      3.75

15:40:26 Sdsk-0     0.07      1.00      0.27      0.67      0.00      2.50

15:40:41 Sdsk-0     1.60      1.00      4.13     10.80      0.00      3.87

15:40:56 Sdsk-0     0.13      1.00      1.27      4.40      0.00      1.05

15:41:11 Sdsk-0     0.40      1.00      1.67      6.93      0.00      2.40
 

The users were now able to agree that the system was faster.



Got something to add? Send me email.





(OLDER)    <- More Stuff -> (NEWER)    (NEWEST)   

Printer Friendly Version

-> -> amird driver, BDFLUSH and NAUTOUP, and more




Increase ad revenue 50-250% with Ezoic


More Articles by

Find me on Google+

© Tony Lawrence




I'm wondering if part of the problem is the LSI controller? I've had mixed results with their RAID adapters, and in fact, seem to recall that they were inferior in terms of short, random access reads to the Mylex units I was using at the time. But, of course, it could be something deep inside the 5.0.7 kernel itself. Who knows?

--BigDumbDinosaur

Kerio Samepage


Have you tried Searching this site?

Unix/Linux/Mac OS X support by phone, email or on-site: Support Rates

This is a Unix/Linux resource website. It contains technical articles about Unix, Linux and general computing related subjects, opinion, news, help files, how-to's, tutorials and more.

Contact us





We are questioning more than the philosophy behind our dependence upon limited and limiting systems. We question the power structures that have grown up around such systems (Frank Herbert).

[C has] the power of assembly language and the convenience of … assembly language. (Dennis Ritchie)












This post tagged: