From: Khosravi, H. M <hor...@in...> - 2004-04-19 18:50:39
|
Hi All BTW, I see an error related to iobw.tst file in /var/log/messages when I run dynamo on my Linux box. Specifically, here is what I see... Automount: attempting to mount entry /net/iobw.tst Automount: >> showmount: cant get address for iobw.tst Automount: Lookup (program): lookup for iobw.tst failed Automount: attempting to mount entry /misc/iobw.tst Automount: Lookup (file): lookup for iobw.tst failed Also, I don't see an 'iobw.tst' file being created anywhere in the system. I have been leaving the 'Maximum Disk Size' value in the IOMeter GUI as 0. Could this be related to the problem ? What should the iobw.tst file look like ?? Thanks Hormuzd -----Original Message----- From: Khosravi, Hormuzd M=20 Sent: Monday, April 19, 2004 10:46 AM To: 'mi...@el...' Cc: 'Dan Bar Dov'; 'iometer-user-list' Subject: RE: [Iometer-user] iometer question I am using Linux kernel... 2.4.20-28.8#1 Thu Dec 18 12:53:39 EST 2003 i686 i686 i386 GNU/Linux I am using IOMeter Version....2003.12.16 Let me know if I should be using different versions for any of these or if there are any known issues ? Hormuzd -----Original Message----- From: Ming Zhang [mailto:mi...@el...]=20 Sent: Monday, April 19, 2004 10:21 AM To: Khosravi, Hormuzd M Cc: Dan Bar Dov; iometer-user-list Subject: RE: [Iometer-user] iometer question Hi, I see. I just jumped into the discussion. :P Can you recompile a vanilla kernel for kernel.org to see if that ok? which kernel from redhat, including all version text here (show me u uname -a). And what iometer version u use? can u try the code from cvs? I know redhat change a lot in kernel and thus the /proc interface may change. in the new iometer code, i already add /proc detect function but i know that can be broken any time. sigh. ming On Mon, 2004-04-19 at 13:02, Khosravi, Hormuzd M wrote: > Hi Ming >=20 > In the results.csv file that is generated by the IOMeter GUI when I run > my tests, there is a column towards the end that says Packet Errors. Its > after Packets/Second. I see lots of packet errors there, which should > not be the case. Most folks who have used IOMeter in my lab have seen 0 > Packet Errors in that column. I suspect this might be related to the > Linux Kernel on my Redhat box. I am running linux 2.4.20, should I be > using a different kernel ? What is the ideal kernel for running dynamo ? >=20 > Thanks > Hormuzd >=20 > -----Original Message----- > From: Ming Zhang [mailto:mi...@el...]=20 > Sent: Monday, April 19, 2004 9:42 AM > To: Khosravi, Hormuzd M > Cc: Dan Bar Dov; iometer-user-list > Subject: RE: [Iometer-user] iometer question >=20 >=20 > Hi, I also use iometer to test iSCSI performance or local disk > performance. No such errors u mentioned. and since u get packet errors > no matter what kind of scsi device u use, so this should have nothing to > do with iscsi. what is u packet error. i mean the original error > message. >=20 > ming >=20 > On Mon, 2004-04-19 at 12:34, Khosravi, Hormuzd M wrote: > > Hi Dan > > =20 > > I am using only devices i.e. /dev/sda, which is the same as what you > > are using. > > =20 > > Thanks > > Hormuzd > > =20 > > -----Original Message----- > > From: Dan Bar Dov [mailto:da...@vo...]=20 > > Sent: Sunday, April 18, 2004 11:47 PM > > To: Khosravi, Hormuzd M; iom...@li... > > Subject: RE: [Iometer-user] iometer question > > =20 > > =20 > > I don't know what packet errors are, on my setup I've never > > seen one (I test iSCSI as well). > > =20 > > Regarding the caching, I did some checks and the O_DIRECT does > > not help, it makes results much lower for some reason I don't > > understand. > > One question though - are you using devices or mounted file > > systems on the iSCSI disk? > > I use only devices (/dev/sda etc.) and never a file system > > since then you get file system caching. > > There is another option with devices that is supposed to > > bypass caching and its using the raw(1) command to link block > > devices such as /dev/sda to character devices like > > /dev/raw/raw1 By using character device you bypass the cache > > buffer, but again, my measurments gave much worse performance > > with that too. > > =20 > > Good luck. > > -----Original Message----- > > From: Khosravi, Hormuzd M > > [mailto:hor...@in...] > > Sent: Sunday, April 18, 2004 10:25 PM > > To: Dan Bar Dov; iom...@li... > > Subject: RE: [Iometer-user] iometer question > > =20 > > =20 > > =20 > > Hi Dan, All > > =20 > > =20 > > =20 > > I believe I have been using a single Worker on the > > iscsi disk. Essentially I run dynamo on my Linux box > > which happens to be the iscsi initiator and have it > > connect to the iometer GUI running on a windows box. > > The Linux box shows up in the iometer GUI and has a > > single worker. I then select the iscsi or non-iscsi > > disks to perform the tests to measure disk i/o. I see > > Packet Errors with both iscsi and non-iscsi disks. > > =20 > > =20 > > =20 > > As regards file caching on Linux, I was told about > > that problem and was advised to use the O_DIRECT flag > > during open() call in the dynamo code (specifically > > IOTargetDisk.cpp) . I tried this but still see > > incorrect nos. Are there any other BKMs to turn off > > file chaching for Linux when using dynamo ? Any ideas > > on why I see Packet Errors ? BTW, I tested my NIC and > > it doesn't seem like having the problem. > > =20 > > =20 > > =20 > > =20 > > =20 > > Thanks for your help > > =20 > > Hormuzd > > =20 > > =20 > > =20 > > =20 > > ______________________________________________________ > > =20 > > From: Dan Bar Dov [mailto:da...@vo...]=20 > > Sent: Sunday, April 18, 2004 3:24 AM > > To: Khosravi, Hormuzd M; > > iom...@li... > > Subject: RE: [Iometer-user] iometer question > > =20 > > =20 > > =20 > > =20 > > I don't know what packet errors are. > > =20 > > =20 > > Regarding throughput, I have found out that if you use > > multiple workers on the same iscsi disk you get > > unreliable bandwidth, probably due to caching effects. > > I don't use more than one worker per iscsi disk in > > order to get reliable results (I verified this wil a > > line monitor). > > =20 > > =20 > > =20 > > =20 > > =20 > > Dan > > =20 > > =20 > > -----Original Message----- > > From: Khosravi, Hormuzd M > > [mailto:hor...@in...] > > Sent: Saturday, April 17, 2004 4:20 AM > > To: iom...@li... > > Subject: RE: [Iometer-user] iometer question > > =20 > > Couple of clarifications regarding my setup... > > =20 > > =20 > > =20 > > =20 > > =20 > > I am running dynamo on a Redhat 8 Linux system > > and IOMeter on Windows based system. > > =20 > > =20 > > Even when I run tests on non-iSCSI disks i.e. > > hard disk, I still see Packet Errors. > > =20 > > =20 > > =20 > > =20 > > =20 > > When I run the tests for iSCSI disks, I see > > Packet Errors and the Throughput is much > > higher that what is possible with a Gig NIC > > connecting the iSCSI client and server (or > > initiator and target) > > =20 > > =20 > > =20 > > =20 > > =20 > > Pls do let me know if you have any idea > > regarding what the problems might be. > > =20 > > =20 > > =20 > > =20 > > =20 > > Thanks > > =20 > > =20 > > Hormuzd > > =20 > > =20 > > -----Original Message----- > > From: > > > iom...@li... > [mailto:iom...@li...] On Behalf Of Khosravi, > Hormuzd M > > Sent: Friday, April 16, 2004 2:24 PM > > To: iom...@li... > > Subject: [Iometer-user] iometer > > question > > =20 > > Hi All > > =20 > > =20 > > =20 > > =20 > > =20 > > I am using iometer to perform some > > tests (on a iSCSI implementation) > > =20 > > =20 > > I see some incorrect results...one > > question is, > > =20 > > =20 > > What does Packet Errors in the > > results.csv stand for ? > > =20 > > =20 > > I see a lot of Packet Errors and that > > might be related to the problem. > > =20 > > =20 > > =20 > > =20 > > =20 > > Thanks a lot, > > =20 > > =20 > > Hormuzd > > =20 |