Subject: Re: VTAPE Performance
From: Kevin Corkery (kcorkery@comcast.net)
Date: Sat Aug 09 2008 - 10:15:00 EDT
Ingo ... I double checked the assembly of $SOCKOPT and it was successfully cataloged into PRD2.CONFIG. This library is the first in the search chain in both the TAPESRVR and TCPIP partitions. The only other copy exisits in PRD1.BASE which is searched after PRD2.CONFIG. The module is not located in the SDL according to LIBR so I'm pretty sure I'm getting the correct version. There's another registry entry (don't have it right in front of me) that will ignore the ACK delay once the queue builds to a certain level but the defualt for it is supposed to be 2 so it would kick in pretty quick making in not very effective unless there's a nuance we're not seeing. If we get a chance to test it and I see an improvement I will let you know. --- Kevin Corkery Independent Consultant Voorhees, New Jersey -------------- Original message ---------------------- From: ifranzki@googlemail.com > Hello Kevin, > > yes, I belive that it makes a big difference. Please see here for info > about how to change the delayed ack settings in Windows: > http://smallvoid.com/article/winnt-nagle-algorithm.html > > However, setting the $OPTSNWT option in $SOCKOPT should have the same > effect. Are you sure that your modified $SOCKOPT is picked up by the > tape server partition ? > > Kind regards > Ingo Franzki, IBM > > On Aug 8, 3:16=A0pm, kcork...@comcast.net (Kevin Corkery) wrote: > > Hi Ingo ... > > > > Well, we found a little ammunition on an Attachmate website which lead us= > to a registry change in Windows that lowered ACK delay from 200ms to 100ms= > (that's as low as it can go) and we saw a dramatic improvement in the spee= > d of the backup. =A0There is a second tweak we found that may also provide = > additional peformance but we're still looking into it. =A0In any case, I ra= > n somewhat equivalent backups on both my Flex system and the z9, faketape v= > ersus vtape, the faketape backup ran about 9 minutes while the vtape backup= > ran about 18 minutes. =A0This is well within our range of expectation. =A0= > The change in settings also resulted in the IP stack being driven at much h= > igher levels on the z9. =A0Before we saw about 6-10% utilization while now = > we're at about 30%, more work to do so more system being used. =A0In all, t= > he performance is now within the range of what we we're hoping to be. > > > > -- > > Kevin Corkery > > Independent Consultant > > Voorhees, New Jersey > >
This archive was generated by hypermail 2b25 : Sat Aug 30 2008 - 00:20:12 EDT