Vdbench, problems with the patterns= parameter

I have written a blog entry about problems with the patterns= parameter before, even mentioning that I may no longer support it. I have concluded since, that I need to continue supporting it though in a different format than currently, where you (in older versions) could specify 127 different data patterns.

In Vdbench 5.01 and and 5.02 (brandnew), patterns= works as follows: patterns=/pattern/dir where file name '/pattern/dir/default' gets picked up, and its contents stored in the data buffer used for writing data. That works.

However, (and these things always happen when it is too late) a few hours after I did the last build of Vdbench 5.02 I realized that yes, I put the pattern in the buffer, but I use the same buffer for reading which means that if you have a mixed read/write workload your data pattern can be overlaid by whatever data is on your disk. Since the pattern is copied only once into your buffer all new writes will NOT contain this pattern. So, until I fix this, if you want a fixed pattern to be written, do not include reads in your test.

In normal operations I use only a single data buffer, both for reads and writes. This is done to save on the amount of memory needed during the run. Loads of luns \* Loads of threads = Loads of memory. This now needs to change when using specific data patterns.

 

Henk.

Comments:

Any help on the following error?

The host is Sun Cluster 3.2 & metasets

Thanks,

[root@host] /opt/swat3 # ./swat
14:51:41.424
14:51:41.427 Tool will expire on: Thu Feb 13 16:45:12 EST 2014
14:51:41.428
14:51:41.925 Swat Version: swat302

14:51:44.078 successfully connected to server localhost
14:51:44.079 Sockets: Connected to server localhost
14:51:44.096 Vsocket.add_client(): apsfiler2-red 14:51:42.865
14:51:44.351 execute(): /usr/bin/iostat -xX
14:51:44.471 execute(): /usr/bin/iostat -xXn
14:51:44.578 execute(): /usr/bin/ls -rlL /dev/rdsk
14:51:45.117 execute(): /opt/swat3/solaris/nicstat 1 999999999
14:51:45.199 execute(): /usr/bin/vmstat 1 999999999
14:51:45.400 starting: /usr/bin/vmstat 1 999999999
14:51:48.499 Local reporting requested; started real time monitor
14:51:50.006 /usr/sbin/metastat -p -s NFS failed
14:51:50.008 /usr/sbin/metastat -p -s NFS: metastat: apsfiler2: NFS: must be owner of the set for this command
14:51:50.008 /usr/sbin/metastat -p -s NFS:
14:51:50.845 /usr/sbin/metastat -p -s samba failed
14:51:50.847 /usr/sbin/metastat -p -s samba: metastat: apsfiler2: samba: must be owner of the set for this command
14:51:50.847 /usr/sbin/metastat -p -s samba:

Posted by Eli Kleinman on February 16, 2010 at 04:55 AM MST #

Swat tries to pick up any available data that can be helpful translating disk names cxtxdxsx to proper SVM, VXVM etc. volume names if applicable. If unsuccessful, Swat will continue normally without SVM or VXVM information.
Henk.

Posted by Henk Vandenbergh on February 16, 2010 at 05:01 AM MST #

Thanks for the help, I wasn't waiting long anof. it now works great.

Posted by Eli Kleinman on February 16, 2010 at 10:32 PM MST #

It's a great tool, especially the replay feature. I have problems using the 'patterns=' parameter, though. Vdbench did not use my pattern in write, even though there was no read at all in the parameter file. I wonder whether you can take a look and see what's wrong with my file (Sorry your header not included):

patterns=L:\\Software\\IO\\VDbench\\vdbench502\\Pattern
fsd=fsd1,anchor=G:/dir1,depth=1,width=1,files=10,size=128k
fwd=fwd1,fsd=fsd1,operation=write,xfersize=64k,fileio=sequential,
fileselect=random,threads=8
rd=rd1,fwd=fwd\*,fwdrate=max,format=yes,elapsed=10,interval=1

Another question: is it possible to specify data pattern for write in replay?

Your help in answering the questions will be very much appreciated.

Thank you.

Posted by Gene Wang on March 11, 2010 at 09:48 AM MST #

Gene,
As you have seen in my blog, there are several problems with the patterns= parameter. These problems will be fixed in vdbench503. If you contact me at vdbench@sun.com I can send you and early copy.
Henk.

Posted by Henk Vandenbergh on March 11, 2010 at 10:05 AM MST #

Post a Comment:
  • HTML Syntax: NOT allowed
About

Blog for Henk Vandenbergh, author of Vdbench, and Sun StorageTek Workload Analysis Tool (Swat). This blog is used to keep you up to date about anything revolving around Swat and Vdbench.

Search

Categories
Archives
« April 2014
SunMonTueWedThuFriSat
  
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
   
       
Today