| Title: | ase |
| Moderator: | SMURF::GROSSO |
| Created: | Thu Jul 29 1993 |
| Last Modified: | Fri Jun 06 1997 |
| Last Successful Update: | Fri Jun 06 1997 |
| Number of topics: | 2114 |
| Total number of notes: | 7347 |
Hi,
As far as I know Informix XPS is a high performance cluster
solution based on a shared nothing model.
It means that each machine on the cluster owns one or more
local disks.
So the database can be spreaded among the several local disks
each of them located and owned by differents machines.
There is no shared disks on shared scsi buses.
A high performance interconnector should be used to link the cluster
members.
There is neither DLM nor DRD mechanisms ton the Informix XPS
solution. That's why memory channel is not even needed. TCR
Production Server and memory channel are not the rigth choice in this
case because they are designed to a shared everything model.
The design of the solution could be something like that :
----------------- -----------------------
| | |
------ ------ ------
| | | | | |
| 1 | | 2 | | 3 |
------ ----- ------
| | |
O.S. disk O.S. disk O.S. disk
disk 1 disk 4 disk 7
disk 2 disk 5 disk 8
disk 3 disk 6 disk 9
The whole database (tables, indexes and so on) is spreaded among the
9 disks (disk1 to 9).
The XPS will know the rigth paths to access the data. If
the data is not local then it will be asked, using XPS mechanisms, to the
other machine that owns it. Example : Machine 1 try to access the disk
7. Disk 7 will always be accessed as a local disk by the machine 3. Then
machine 3 sends the data back to machine 1.
So I have 3 questions :
1) First of all I would like to know if my concepts about the Informix
XPS are correct ?
2) Can I use the Trucluster Memory Channel Software product to support
the Informix XPS solution ?
3) If the database is spreaded among several disks located and owned by
different machines and being accessed by local buses instead of shared
buses what happens in case of failure of on of the cluster members ?
Need I to replicate data contained on disk 1 to others disks (such as
disk 4 and disk 7)? How can the XPS still access the disk 7 data in case of
failure of the machine 3?
Best regards,
Leo
Digital Technical Support
Brazil
| T.R | Title | User | Personal Name | Date | Lines |
|---|---|---|---|---|---|
| 1985.1 | KITCHE::schott | Eric R. Schott USG Product Management | Sun Apr 06 1997 23:50 | 104 | |
> > As far as I know Informix XPS is a high performance cluster > solution based on a shared nothing model. > > It means that each machine on the cluster owns one or more > local disks. > > So the database can be spreaded among the several local disks > each of them located and owned by differents machines. > > There is no shared disks on shared scsi buses. XPS wants shared disks to do failover...the current release does not support failover, but they will in the next couple months. > > > A high performance interconnector should be used to link the cluster > members. Yup...on Digital UNIX this is MEMORY CHANNEL. > > There is neither DLM nor DRD mechanisms ton the Informix XPS > solution. That's why memory channel is not even needed. TCR > Production Server and memory channel are not the rigth choice in this > case because they are designed to a shared everything model. Huh? You need a high performance interconnect... You want failover (once XPS supports it) You want shared buses (for failover). In short, you need the productions server. > > The design of the solution could be something like that : > > ----------------- ----------------------- > | | | > ------ ------ ------ > | | | | | | > | 1 | | 2 | | 3 | > ------ ----- ------ > | | | > O.S. disk O.S. disk O.S. disk > disk 1 disk 4 disk 7 > disk 2 disk 5 disk 8 > disk 3 disk 6 disk 9 > > The whole database (tables, indexes and so on) is spreaded among the > 9 disks (disk1 to 9). > > The XPS will know the rigth paths to access the data. If > the data is not local then it will be asked, using XPS mechanisms, to the > other machine that owns it. Example : Machine 1 try to access the disk > 7. Disk 7 will always be accessed as a local disk by the machine 3. Then > machine 3 sends the data back to machine 1. Basically this is correct, but you picture is wrong...you need a high speed interconnect, and you need shared storage...you don't show both of these. > > > So I have 3 questions : > > 1) First of all I would like to know if my concepts about the Informix > XPS are correct ? See above...yes and no... > > 2) Can I use the Trucluster Memory Channel Software product to support > the Informix XPS solution ? Nope, production server only. > > 3) If the database is spreaded among several disks located and owned by > different machines and being accessed by local buses instead of shared > buses what happens in case of failure of on of the cluster members ? > Need I to replicate data contained on disk 1 to others disks (such as > disk 4 and disk 7)? How can the XPS still access the disk 7 data in case of > failure of the machine 3? Currently XPS does not support failover, so you are down...in the future when they support failover, you need shared buses... > > > Best regards, > > > Leo > Digital Technical Support > Brazil > > > > | |||||