HDR performance to remote site with IDS 10

Hello, I have configured HDR several times from online 7 through IDS 10 but always over a LAN with both hosts at the same location. We have tried our first IDS HDR pair over a remote link with both servers connected over a 2Mb/s link with a latency of around 10ms. Replication is set asynchronously. Data is being loaded into few tables only from an old SE db that the new instance replaces, this is a simple SQL script loading a small number of .unl files. This data size is around 13Mb. With the primary in standalone mode the load is very fast, when HRD is active loading the same tables is very slow. Varying DRINREVAL also varies the load time as follows. Load times; standalone 00:00:13 DRINTERVAL = 30 00:05:12 DRINTERVAL = 300 00:04:48 So my question to you all is does this seem normal? Chunks size free bpages flags pathname 1048500 1045658 PO-B /dev/informix/rootdbs 1048576 23 PO-B /dev/informix/physicallogdbs 2621440 1621387 PO-B /dev/informix/logicallogsdbs 262144 241160 PO-B /dev/informix/nmcdbs Regards Mike

MIKE SLAUGHTER said: > > We have tried our first IDS HDR pair over a remote link with both servers > connected over a 2Mb/s link with a latency of around 10ms. Replication is > set > asynchronously. Data is being loaded into few tables only from an old SE > db > that the new instance replaces, this is a simple SQL script loading a > small > number of .unl files. This data size is around 13Mb. > > With the primary in standalone mode the load is very fast, when HRD is > active > loading the same tables is very slow. Varying DRINREVAL also varies the > load > time as follows. I don't have anything useful to add to this, other than I hope Madison is keeping his eye on this, because this kind of thing is going to become much more prevalent over time. :o( -- Bye now, Obnoxio "I'm astonished anyone pays real money for this crap." -- Cosmo -- This message has been scanned for viruses and dangerous content by OpenProtect(http://www.openprotect.com), and is believed to be clean.

I've indirectly addressed this problem before. HDR uses half-duplexed network protocol. That means that before the ne= xt log buffer is sent that an ACK must be received from the last buffer se= nt. There are several reasons for this, but it is a requirement for some of= the things that HDR supports. That means that HDR is a bit sensitive to sluggish networks. Since when all of the HDR send buffers are full and= waiting to be transmitted, the logging system can not flush the log buf= fers as fast. In effect, the logging system will have to wait until there i= s a free HDR send buffer before it can release the log buffer which it is trying to flush. Basically if the ping time between the primary and the secondary is 0.3= 33 of a second, then that means that you will not be able to perform much = more than 2-3 log flushes per second - which will impact performance on the primary node. We have addressed this somewhat in IDS11 with the RSS node by using a f= ully duplexed communication between the primary and the RSS node. This allo= ws us to 'pack' the network from the primary to the secondary and to mini= mize the number of ACKs being returned from the secondary back to the primar= y. That means better utilization of the bandwidth - which should make RSS = have less impact on the primary and also be a bit better at keeping up. Th= ere are currently some restrictions on what the RSS node can support, but I= suspect that we may have to improve a bit on those restrictions in upco= ming releases because I suspect that more folks are going to switch to a WAN= for availability support. For the existing pre-IDS11 customer, probably the best I can suggest is= to increase the log buffer size so that more data will be sent per transmi= t. M.P. = "MIKE SLAUGHTER" = <mike.slaughter@c = ognitomobile.com> = To Sent by: ids@iiug.org = ids-bounces@iiug. = cc org = Subj= ect HDR performance to remote site w= ith 06/21/2007 04:27 IDS 10 [9398] = AM = = = Please respond to = ids@iiug.org = = = Hello, I have configured HDR several times from online 7 through IDS 10 but al= ways over a LAN with both hosts at the same location. We have tried our first IDS HDR pair over a remote link with both serve= rs connected over a 2Mb/s link with a latency of around 10ms. Replication = is set asynchronously. Data is being loaded into few tables only from an old S= E db that the new instance replaces, this is a simple SQL script loading a s= mall number of .unl files. This data size is around 13Mb. With the primary in standalone mode the load is very fast, when HRD is active loading the same tables is very slow. Varying DRINREVAL also varies the= load time as follows. Load times; standalone 00:00:13 DRINTERVAL =3D 30 00:05:12 DRINTERVAL =3D 300 00:04:48 So my question to you all is does this seem normal? Chunks size free bpages flags pathname 1048500 1045658 PO-B /dev/informix/rootdbs 1048576 23 PO-B /dev/informix/physicallogdbs 2621440 1621387 PO-B /dev/informix/logicallogsdbs 262144 241160 PO-B /dev/informix/nmcdbs Regards Mike ***********************************************************************= ******** Forum Note: Use "Reply" to post a response in the discussion forum. =

For us, it almost seems quicker (as long as your secondary is just a standby) to break HDR, do the loading/alters/inserts/indexing on the primary and then restart the secondary. The log shipping and syncing usually flies along pretty good. Bob Roussey Unix / Informix Administration Spirit Airlines Robert.Roussey@SpiritAir.com -----Original Message----- From: ids-bounces@iiug.org [mailto:ids-bounces@iiug.org] On Behalf Of MIKE SLAUGHTER Sent: Thursday, June 21, 2007 5:27 AM To: ids@iiug.org Subject: HDR performance to remote site with IDS 10 [9398] Hello, I have configured HDR several times from online 7 through IDS 10 but always over a LAN with both hosts at the same location. We have tried our first IDS HDR pair over a remote link with both servers connected over a 2Mb/s link with a latency of around 10ms. Replication is set asynchronously. Data is being loaded into few tables only from an old SE db that the new instance replaces, this is a simple SQL script loading a small number of .unl files. This data size is around 13Mb. With the primary in standalone mode the load is very fast, when HRD is active loading the same tables is very slow. Varying DRINREVAL also varies the load time as follows. Load times; standalone 00:00:13 DRINTERVAL = 30 00:05:12 DRINTERVAL = 300 00:04:48 So my question to you all is does this seem normal? Chunks size free bpages flags pathname 1048500 1045658 PO-B /dev/informix/rootdbs 1048576 23 PO-B /dev/informix/physicallogdbs 2621440 1621387 PO-B /dev/informix/logicallogsdbs 262144 241160 PO-B /dev/informix/nmcdbs Regards Mike ************************************************************************ ******* Forum Note: Use "Reply" to post a response in the discussion forum.