Re: 2xSMP vs RAC for failover - was Re: informix market share

DA Morgan said: > > Serge Rielau wrote: >> DA Morgan wrote: >> >>> Serge Rielau wrote: >>> >>>> *chuckle*And there always will be because SMP scales better than RAC >>>> (more CPUs needed) and RAC is cited as a major contributer to Linux >>>> licenses... it ain't free. >>> >>> >>> >>> Of course if you also want transparent failover with that big SMP box >>> you get to buy at least one more of them. >> >> Of course. But that wou;d be the add Orcale doen't show. > > Actually that IS the Oracle marketing message. Stop buying big, > expensive, SMP boxes. Instead buy low cost resilient hardware. Build > the equivalent of those expensive boat-anchors from components and > gain not just equivalent processing power and lower cost but, and a > most important ... but ... also gain transparent failover. > >>> From my experience only a small percentage of Oracle customers buy RAC >>> for scalability. The vast majority for transparent failover. >> >> That matches what I hear. Not what Oracle says of course (that ad >> again).... > > Any technologist that listens to marketing messages and makes the > assumption that they are the definitive word is a fool. I would hope one > does not spend a decade in this industry only to swoon over the > carefully crafted verbiage of some spin doctor. > >>> Want to >>> cost it out against alternative technologies with equivalent features? >>> Oh and without having to manually federate the data? Wouldn't want you >>> to forget that part. >> >> Actually yes. Pricing out a failover solution would be fun. >> So we're talking 2 SMP boxes with failover vs. the equivalent commondity >> RAC cluster, including switches, disk failover, software. > > I've done it numerous times and also use it as an exercise for my > advanced students. The price differential is amazingly large. > >> I'm no expert in failover. Can I presume the core contenders are IDS >> with HDR (this is an Informix/Oracle thread), Dataguard and RAC (what >> about Veritas?) > > Veritas, at least from an Oracle standpoint, is irrelevant. > > The Oracle technologies would be RAC ... for failover within a cluster, > DataGuard for failover between clusters ... RMAN for the ability to do > single block recovery ... and ASM for mirroring. > >> Of course without federation. Federation is for scale out which we just >> took of the table. > > No. I took federation off the table because DB2 federates data on all of > its systems except the mainframe and I didn't want a mainframe solution > put up as a contender against Linux/UNIX solutions. Though given the > cost of the hardware, maintenance agreements, and operators from a > financial perspective it would make a juicy comparison. > >> I'm no hardware guy, but I take it the group will keep us honest. >> And I truly want to learn this stuff. >> >> I took the liberty to rename the thread. >> >> Cheers >> Serge > > So here's what I get when I do it ... looking at just the hardware, > operating system, and maintenance agreement cost. > > I can build a 2-node, 4-CPU RAC cluster including 3 years of 7x24 > maintenance, including rail kits for under $6K USD. And with that, > using Oracle Standard Edition, I get transparent application fail- > over (RAC and TAF) included. > > Now price it out with 4-CPU boxes ... well if you want failover you > need to buy two of them. So now you price out 2x4-CPU boxes with > 7x24 maintenance and rail kits and tell me if you can beat $6K USD. > And 1/2 of that CPU power and 1/2 of that investment is sitting > idle in the corner in case something goes terribly wrong with the > first expensive box. > > I currently have 2 x 10-node clusters (two 20CPU RAC clusters) > operating in Japan one at the main data center another at a remote > backup location. Cost of hardware ... $60K USD for all 40 CPUs. > Compare that $60K USD against the cost of two 20 CPU boxes. And > feel free to round down in the thousands place. Have you tried UPDATE STATISTICS? Serge, don't you please want to take this Oracle vs DB2 pissing contest somewhere else? I'm asking you, because I know Daniel doesn't care where he waves his cock. -- Bye now, Obnoxio "C'est pas parce qu'on n'a rien ` dire qu'il faut fermer sa gueule" - Coluche did i mention i like nulls? heck, i even go so far as to say that all columns in a table except the primary key could/should be nullable. this has certain advantages, for example, if you need to insert a child record and you don't have a parent row for it, just do an insert into the parent table with the primary key value (everything else null), and voila, relational integrity is preserved. but this is, admittedly, a bit controversial among modellers. --r937, dbforums.com sending to informix-list

Obnoxio The Clown wrote: > DA Morgan said: > >>Serge Rielau wrote: >> >>>DA Morgan wrote: >>> >>> >>>>Serge Rielau wrote: >>>> >>>> >>>>>*chuckle*And there always will be because SMP scales better than RAC >>>>>(more CPUs needed) and RAC is cited as a major contributer to Linux >>>>>licenses... it ain't free. >>>> >>>> >>>> >>>>Of course if you also want transparent failover with that big SMP box >>>>you get to buy at least one more of them. >>> >>>Of course. But that wou;d be the add Orcale doen't show. >> >>Actually that IS the Oracle marketing message. Stop buying big, >>expensive, SMP boxes. Instead buy low cost resilient hardware. Build >>the equivalent of those expensive boat-anchors from components and >>gain not just equivalent processing power and lower cost but, and a >>most important ... but ... also gain transparent failover. >> >> >>>>From my experience only a small percentage of Oracle customers buy RAC >>>>for scalability. The vast majority for transparent failover. >>> >>>That matches what I hear. Not what Oracle says of course (that ad >>>again).... >> >>Any technologist that listens to marketing messages and makes the >>assumption that they are the definitive word is a fool. I would hope one >>does not spend a decade in this industry only to swoon over the >>carefully crafted verbiage of some spin doctor. >> >> >>>>Want to >>>>cost it out against alternative technologies with equivalent features? >>>>Oh and without having to manually federate the data? Wouldn't want you >>>>to forget that part. >>> >>>Actually yes. Pricing out a failover solution would be fun. >>>So we're talking 2 SMP boxes with failover vs. the equivalent commondity >>>RAC cluster, including switches, disk failover, software. >> >>I've done it numerous times and also use it as an exercise for my >>advanced students. The price differential is amazingly large. >> >> >>>I'm no expert in failover. Can I presume the core contenders are IDS >>>with HDR (this is an Informix/Oracle thread), Dataguard and RAC (what >>>about Veritas?) >> >>Veritas, at least from an Oracle standpoint, is irrelevant. >> >>The Oracle technologies would be RAC ... for failover within a cluster, >>DataGuard for failover between clusters ... RMAN for the ability to do >>single block recovery ... and ASM for mirroring. >> >> >>>Of course without federation. Federation is for scale out which we just >>>took of the table. >> >>No. I took federation off the table because DB2 federates data on all of >>its systems except the mainframe and I didn't want a mainframe solution >>put up as a contender against Linux/UNIX solutions. Though given the >>cost of the hardware, maintenance agreements, and operators from a >>financial perspective it would make a juicy comparison. >> >> >>>I'm no hardware guy, but I take it the group will keep us honest. >>>And I truly want to learn this stuff. >>> >>>I took the liberty to rename the thread. >>> >>>Cheers >>>Serge >> >>So here's what I get when I do it ... looking at just the hardware, >>operating system, and maintenance agreement cost. >> >>I can build a 2-node, 4-CPU RAC cluster including 3 years of 7x24 >>maintenance, including rail kits for under $6K USD. And with that, >>using Oracle Standard Edition, I get transparent application fail- >>over (RAC and TAF) included. >> >>Now price it out with 4-CPU boxes ... well if you want failover you >>need to buy two of them. So now you price out 2x4-CPU boxes with >>7x24 maintenance and rail kits and tell me if you can beat $6K USD. >>And 1/2 of that CPU power and 1/2 of that investment is sitting >>idle in the corner in case something goes terribly wrong with the >>first expensive box. >> >>I currently have 2 x 10-node clusters (two 20CPU RAC clusters) >>operating in Japan one at the main data center another at a remote >>backup location. Cost of hardware ... $60K USD for all 40 CPUs. >>Compare that $60K USD against the cost of two 20 CPU boxes. And >>feel free to round down in the thousands place. > > > Have you tried UPDATE STATISTICS? > > Serge, don't you please want to take this Oracle vs DB2 pissing contest > somewhere else? I'm asking you, because I know Daniel doesn't care where > he waves his cock. > Duly noted. Thoughts for the road: If you use a 2 node cluster and one goes down you're dead unless you kept load very low when the RAC system is up. That is the 2 node RAC cluster competes with a 2x4 SMP box which of course will price hardware wise the same (minus the 2 switches + disk). You did not price out the equivalent dataguard solution, btw. Cheers and bye Serge -- Serge Rielau DB2 SQL Compiler Development IBM Toronto Lab