Re: HiperSockets Question
On Wednesday, 08/20/2008 at 09:54 EDT, Martin, Terry R. (CMS/CTR) (CTR) [EMAIL PROTECTED] wrote: After a lot of effort we found that there was an extra route in the Linux routing table that sent the data bound for one of the z/OS LPARS over to our z/VM where it then was routed by the TCP/IP stack on z/VM to the z/OS LPAR over the HiperSockets LINK. This LINK happened to have a MTU size of 1500. When we removed the bogus route from the Linux routing table causing the SQL queries in question to travel over a HiperSockets LINK with a MTU size of 16348 they were successful. It appears that the difference in the MTU sizes was the problem. My question is does the MTU size mismatch cause the problem? I thought that even if the packets were fragmented they would still get over in full. Yes, it can (but not always) cause a problem if the MTU sizes on the same shared media are different. All MTUs on the same network segment need to have the same MTU. Alan Altmark z/VM Development IBM Endicott
Re: HiperSockets Question
Thanks Alan! Thank You, Terry Martin Lockheed Martin - Information Technology z/OS z/VM Systems - Performance and Tuning Cell - 443 632-4191 Work - 410 786-0386 [EMAIL PROTECTED] -Original Message- From: The IBM z/VM Operating System [mailto:[EMAIL PROTECTED] On Behalf Of Alan Altmark Sent: Friday, August 22, 2008 9:11 AM To: IBMVM@LISTSERV.UARK.EDU Subject: Re: HiperSockets Question On Wednesday, 08/20/2008 at 09:54 EDT, Martin, Terry R. (CMS/CTR) (CTR) [EMAIL PROTECTED] wrote: After a lot of effort we found that there was an extra route in the Linux routing table that sent the data bound for one of the z/OS LPARS over to our z/VM where it then was routed by the TCP/IP stack on z/VM to the z/OS LPAR over the HiperSockets LINK. This LINK happened to have a MTU size of 1500. When we removed the bogus route from the Linux routing table causing the SQL queries in question to travel over a HiperSockets LINK with a MTU size of 16348 they were successful. It appears that the difference in the MTU sizes was the problem. My question is does the MTU size mismatch cause the problem? I thought that even if the packets were fragmented they would still get over in full. Yes, it can (but not always) cause a problem if the MTU sizes on the same shared media are different. All MTUs on the same network segment need to have the same MTU. Alan Altmark z/VM Development IBM Endicott
HiperSockets Question
Hi Alan, I thought I sent this out last night but it does not look like it went so I am sending it again. If for some reason you did get this I apologize for sending it again. Not to belabor the subject I just want to make sure I understand. Here is what my issue was. I processed many SQL queries of DB2 tables via DB2 connect running on a z/OS LPAR down to a Linux server running on the same machine but a different z/VM LPAR without any issues. Until, we ran across some queries that were larger than all of the others. These queries failed (timed out). We saw in the DB2 connect trace that the size of the queries was about 2.3K. We knew it was a size issue because when we did a 'SELECT *' on the tables, basically reducing the size it worked. After a lot of effort we found that there was an extra route in the Linux routing table that sent the data bound for one of the z/OS LPARS over to our z/VM where it then was routed by the TCP/IP stack on z/VM to the z/OS LPAR over the HiperSockets LINK. This LINK happened to have a MTU size of 1500. When we removed the bogus route from the Linux routing table causing the SQL queries in question to travel over a HiperSockets LINK with a MTU size of 16348 they were successful. It appears that the difference in the MTU sizes was the problem. My question is does the MTU size mismatch cause the problem? I thought that even if the packets were fragmented they would still get over in full. Sorry for being so wordy but I thought I would present to try to help you help me and to also help others that may be just getting into this as I am. Thanks again for all of your help!! Terry Thank You, Terry Martin Lockheed Martin - Information Technology z/OS z/VM Systems - Performance and Tuning Cell - 443 632-4191 Work - 410 786-0386 [EMAIL PROTECTED]
HiperSockets Question
Hi If I am using HiperSockets and my MTU size is 1500 and a pass a packet size of 2000 will the data in the packet be truncated since it is bigger than 1500? I thought I read that since HiperSockets is memory transfers that the MTU can not be split like it would over a normal transmission pipe, meaning that for a 3K packet size over a MTU 1500 link would not be split into 2 transmissions. Am I correct in the assumption? Thank You, Terry Martin Lockheed Martin - Information Technology z/OS z/VM Systems - Performance and Tuning Cell - 443 632-4191 Work - 410 786-0386 [EMAIL PROTECTED]
Re: HiperSockets Question
On Wednesday, 08/20/2008 at 03:43 EDT, Martin, Terry R. (CMS/CTR) (CTR) [EMAIL PROTECTED] wrote: If I am using HiperSockets and my MTU size is 1500 and a pass a packet size of 2000 will the data in the packet be truncated since it is bigger than 1500? I thought I read that since HiperSockets is memory transfers that the MTU can not be split like it would over a normal transmission pipe, meaning that for a 3K packet size over a MTU 1500 link would not be split into 2 transmissions. Am I correct in the assumption? Yes and no. :-) It won't be truncated, but it will be fragmented. The MTU is the Guardian of the Gate. It prevents the IP layer from sending a packet too large for the underlying interface to handle. So a 3K packet over an MTU 1500 link will result in two fragments, each 1500 bytes. Alan Altmark z/VM Development IBM Endicott
Re: HiperSockets Question
Packet fragmentation is done in the TCP/IP stack. The stack will chunk the packet up into smaller packets before sending it over the HiperSockets device. Regards, Miguel Delapaz z/VM TCP/IP Development The IBM z/VM Operating System IBMVM@LISTSERV.UARK.EDU wrote on 08/20/2008 12:40:47 PM: [image removed] HiperSockets Question Martin, Terry R. (CMS/CTR) (CTR) to: IBMVM 08/20/2008 12:43 PM Sent by: The IBM z/VM Operating System IBMVM@LISTSERV.UARK.EDU Please respond to The IBM z/VM Operating System Hi If I am using HiperSockets and my MTU size is 1500 and a pass a packet size of 2000 will the data in the packet be truncated since it is bigger than 1500? I thought I read that since HiperSockets is memory transfers that the MTU can not be split like it would over a normal transmission pipe, meaning that for a 3K packet size over a MTU 1500 link would not be split into 2 transmissions. Am I correct in the assumption? Thank You, Terry Martin Lockheed Martin - Information Technology z/OS z/VM Systems - Performance and Tuning Cell - 443 632-4191 Work - 410 786-0386 [EMAIL PROTECTED]
Re: HiperSockets Question
If you wanted to get nitpicky, it would actually need 3 fragments to account for the extra IP header(s). :-) Regards, Miguel Delapaz z/VM TCP/IP Development The IBM z/VM Operating System IBMVM@LISTSERV.UARK.EDU wrote on 08/20/2008 01:38:48 PM: Yes and no. :-) It won't be truncated, but it will be fragmented. The MTU is the Guardian of the Gate. It prevents the IP layer from sending a packet too large for the underlying interface to handle. So a 3K packet over an MTU 1500 link will result in two fragments, each 1500 bytes. Alan Altmark z/VM Development IBM Endicott
Re: HiperSockets Question
On Wednesday, 08/20/2008 at 04:44 EDT, Miguel Delapaz/Endicott/[EMAIL PROTECTED] wrote: If you wanted to get nitpicky, it would actually need 3 fragments to account for the extra IP header(s). :-) I *knew* I'd regret not diving into nitty gritty details. I KNEW IT. Mr. Smartypants. Couldn't let it go, could you? We were in the middle of a *point*. It's going to cost you. Not today, not tomorrow, but when you least expect it. No need to look over your shoulder - you won't see me. ;-) Sleep well (but not too deeply). As Khan Noonien Singh said, Revenge is a dish best served cold. Alan da Weasel
Re: HiperSockets Question
At 07:09 PM 8/20/2008 -0400, Alan Altmark wrote: I *knew* I'd regret not diving into nitty gritty details. I KNEW IT. Mr. Smartypants. Couldn't let it go, could you? We were in the middle of a *point*. DTCMPY001S Nobody ever expects the Spanish Inquisition.
Re: HiperSockets Question
Isn't that Sir Alan da Weasel? :) Alan Altmark wrote: Alan da Weasel -- Rich Smrcina VM Assist, Inc. Phone: 414-491-6001 Ans Service: 360-715-2467 rich.smrcina at vmassist.com http://www.linkedin.com/in/richsmrcina Catch the WAVV! http://www.wavv.org WAVV 2009 - Orlando, FL - May 15-19, 2009
Re: HiperSockets Question
On Wednesday, 08/20/2008 at 07:51 EDT, Rich Smrcina [EMAIL PROTECTED] wrote: Isn't that Sir Alan da Weasel? :) Please. Was Zorro, Don Zorro? I think not. Alan da Weasel
Re: HiperSockets Question
He was El Zorro. And Quickdraw McGraw was El Kabong. The 'El' does make a difference. /Tom Kern Alan Altmark wrote: On Wednesday, 08/20/2008 at 07:51 EDT, Rich Smrcina [EMAIL PROTECTED] wrote: Isn't that Sir Alan da Weasel? :) Please. Was Zorro, Don Zorro? I think not. Alan da Weasel
Re: HiperSockets Question
Hi Alan, Not to belabor the subject I just want to make sure I understand. Here is what my issue was. I processed many SQL queries of DB2 tables via DB2 connect running on a z/OS LPAR down to a Linux server running on the same machine but a different z/VM LPAR without any issues. Until, we ran across some queries that were larger than all of the others. These queries failed (timed out). We saw in the DB2 connect trace that the size of the queries was about 2.3K. We knew it was a size issue because when we did a 'SELECT *' on the tables, basically reducing the size it worked. After a lot of effort we found that there was an extra route in the Linux routing table that sent the data bound for one of the z/OS LPARS over to our z/VM where it then was routed by the TCP/IP stack on z/VM to the z/OS LPAR over the HiperSockets LINK. This LINK happened to have a MTU size of 1500. When we removed the bogus route from the Linux routing table causing the SQL queries in question to travel over a HiperSockets LINK with a MTU size of 16348 they were successful. It appears that the difference in the MTU sizes was the problem. My question is does the MTU size mismatch cause the problem? I thought that even if the packets were fragmented they would still get over in full. Sorry for being so wordy but I thought I would present to try to help you help me and to also help others that may be just getting into this as I am. Thanks again for all of your help!! Terry -Original Message- From: The IBM z/VM Operating System [mailto:[EMAIL PROTECTED] On Behalf Of Alan Altmark Sent: Wednesday, August 20, 2008 4:39 PM To: IBMVM@LISTSERV.UARK.EDU Subject: Re: HiperSockets Question On Wednesday, 08/20/2008 at 03:43 EDT, Martin, Terry R. (CMS/CTR) (CTR) [EMAIL PROTECTED] wrote: If I am using HiperSockets and my MTU size is 1500 and a pass a packet size of 2000 will the data in the packet be truncated since it is bigger than 1500? I thought I read that since HiperSockets is memory transfers that the MTU can not be split like it would over a normal transmission pipe, meaning that for a 3K packet size over a MTU 1500 link would not be split into 2 transmissions. Am I correct in the assumption? Yes and no. :-) It won't be truncated, but it will be fragmented. The MTU is the Guardian of the Gate. It prevents the IP layer from sending a packet too large for the underlying interface to handle. So a 3K packet over an MTU 1500 link will result in two fragments, each 1500 bytes. Alan Altmark z/VM Development IBM Endicott
