Proxmox Newbie - Running 2 Physical Servers and 2 VMs - Roadmap Issue

[Hobbystern]

Hello Guys,

i am new to this forum, but not new to linux and virtualization.

I planned more than one year the step to consolidate my servers in the house. The Goal is to get 2 identically Servers (actually 2 HP ProLiant DL385 with 10GB RAM and 73GB for Proxmox and Backups (RAID1) , 300GB for VMs (RAID5) - redundant)

I have :

1 Asterisk Server (Debian, with a professional Asterisk Implementation and more than 50 Clients) - no much data (4GB) but many configuration files and fine tuning - i would be very happy if i could "convert" this physical machine directly to a vm (openvz)

1 Samba and VM Server (Debian, too. Samba 3.x, DNS, Mailgateway, BackupPC, Virtualbox (not very good for high availability) and 1 windows 2008 server with a very good protected MS SQL Server 2005 (company data on it-must be ha!) running inside the vm. This Server got much SATA Space (1 TB with RAID10) which CANT be used any more, or i share it as a NFS Server. 1 old Windows XP "Testrange" in a virtual enviroment.

My absolutely first goal is to get the system redundant, same data on the first as on the second server, using heartbeat (or as i see on the roadmap corosync) with drbd is very recommend. This Goal implements the need of ONE Master and ONE idle Slave - as i can see on the roadmap and read before in several books, coronet implements the MASTER/MASTER ability. This would be very good - but i have to take care about my databases on the 2008Server-VM (broken data is not my plan ) - my hp proliant server are with ILO, so i can use stonith to get this situation under control.

My Questions :

--> Is it good to migrate all this NOW (with knowing that corosync will come in 2.x and will change everything)
--> Is it possible to migrate a Debian Server (i only found "migration of windows systems" in the wiki and the howtos)
--> Is it actually possible to build a HA System without doing everything by hand (its not a problem, but i keep in my mind - if something went wrong and one system is going down, so everything must be restored to the old "hand" state - this is a case that i dont want on my proxmox ve base-system - i was very happy to see how short takes a complete new installation from ISO. Configuration takes me 20 minutes and thats it. Best case!

At this place i must say "Thanks ! For Proxmox VE - looks like a good idea to set it on a debian system, as you can see - as many other people, i like debian.)

Thanks for your time if you have read my post to this point. Thanks for any helpful comment!

Regards! Hobbystern (Germany)

 

[udo]

Hello Guys,

i am new to this forum, but not new to linux and virtualization.

I planned more than one year the step to consolidate my servers in the house. The Goal is to get 2 identically Servers (actually 2 HP ProLiant DL385 with 10GB RAM and 73GB for Proxmox and Backups (RAID1) , 300GB for VMs (RAID5) - redundant)

I have :

1 Asterisk Server (Debian, with a professional Asterisk Implementation and more than 50 Clients) - no much data (4GB) but many configuration files and fine tuning - i would be very happy if i could "convert" this physical machine directly to a vm (openvz)

Hi,
directly convert isn't possible with openvz - you must transfer your services. Therefore you get an lightweigt system. But openvz needs local storage - not realy ha!
And for openvz it's the best to use the 2.6.18-kernel (many nice things need 2.6.32 (or .35)).

1 Samba and VM Server (Debian, too. Samba 3.x, DNS, Mailgateway, BackupPC, Virtualbox (not very good for high availability) and 1 windows 2008 server with a very good protected MS SQL Server 2005 (company data on it-must be ha!) running inside the vm. This Server got much SATA Space (1 TB with RAID10) which CANT be used any more, or i share it as a NFS Server. 1 old Windows XP "Testrange" in a virtual enviroment.

My absolutely first goal is to get the system redundant, same data on the first as on the second server, using heartbeat (or as i see on the roadmap corosync) with drbd is very recommend. This Goal implements the need of ONE Master and ONE idle Slave - as i can see on the roadmap and read before in several books, coronet implements the MASTER/MASTER ability. This would be very good - but i have to take care about my databases on the 2008Server-VM (broken data is not my plan ) - my hp proliant server are with ILO, so i can use stonith to get this situation under control.

My Questions :

--> Is it good to migrate all this NOW (with knowing that corosync will come in 2.x and will change everything)
--> Is it possible to migrate a Debian Server (i only found "migration of windows systems" in the wiki and the howtos)

the reason, why you only found windows migration is that windows makes troubel. A linux-box migration is normaly no problem (copy raw-image; boot and perhaps remove old udev-entry of old nic).

--> Is it actually possible to build a HA System without doing everything by hand (its not a problem, but i keep in my mind - if something went wrong and one system is going down, so everything must be restored to the old "hand" state - this is a case that i dont want on my proxmox ve base-system - i was very happy to see how short takes a complete new installation from ISO. Configuration takes me 20 minutes and thats it. Best case!

No, thats not realy possible. You can start (also with scripting) an machine on the other server, but this is like a powercycle. The memory/cpu-state are not in sync (i'm think this will be also not with pve2.0).

At this place i must say "Thanks ! For Proxmox VE - looks like a good idea to set it on a debian system, as you can see - as many other people, i like debian.)

Thanks for your time if you have read my post to this point. Thanks for any helpful comment!

Regards! Hobbystern (Germany)

You should test your shared storage. If you use FC-/iScsi-Raids you have a spof, this isn't with drbd. But check the speed (latency, throughput) for all solutions.

Udo

 

[Hobbystern]

Hey Udo,

first i have to say thanks for your reply.

But openvz needs local storage - not realy ha!

OpenVZ and a separate Partition, storing all the vm-data and mem states - deployed with drbd will eliminate this problem?

And for openvz it's the best to use the 2.6.18-kernel (many nice things need 2.6.32 (or .35)).

Ah. ok. My productionally systems using 2.6.26 (bigmem) and the last time i seen 2.6.18 is ..... 3 years ago? I cant use KVM :-( because my proliant servers are fast, but not with amd-v equipped.

that windows makes troubel.

without any words

The memory/cpu-state are not in sync (i'm think this will be also not with pve2.0).

In my opinion for a Master/Slave System the first thing to do is to keep the stored data the same (drbd or a shared drive, iscsi...) - the usage is step 2 - i think. It is okay to stay in a M/S State, because my goal is high availability. its nice if it goes to master/master state, but only nice.

You should test your shared storage. If you use FC-/iScsi-Raids you have a spof, this isn't with drbd. But check the speed (latency, throughput) for all solutions.

The projected "miniplan" of consolidating my servers keeps care of taking local space (raid1 and redundant with drbd (protocol B) for ha-vms. All the "not ha needed" vms can take place on shared drives - shared drives will ever be a spof detail in my opinion.

What do you think - is proxmox ha ready at this time or should i build up my own ha systems - as i planned in the first step, before knowing proxmox.

Stefan!

 

[udo]

Hey Udo,

first i have to say thanks for your reply.

OpenVZ and a separate Partition, storing all the vm-data and mem states - deployed with drbd will eliminate this problem?

Hi Stefan,
perhaps - but this will need some (more) handwork. AFAIK you can't use the livemigration with proxmox, because proxmox rsync the content of the vm (this is perhaps no problem) and remove after that the old one (and with drdb both sides are gone...)

Ah. ok. My productionally systems using 2.6.26 (bigmem) and the last time i seen 2.6.18 is ..... 3 years ago? I cant use KVM :-( because my proliant servers are fast, but not with amd-v equipped.

the proxmox-team backportet a lot of driver to the 2.6.18 - so this kernel is much "newer" - For OpenVZ it's at this time the only rock-stable kernel! And if you have proliant-server without hw-virtualization, the 2.6.18 is new enough ( i had also such an server - other distros are not installable ;-( )

In my opinion for a Master/Slave System the first thing to do is to keep the stored data the same (drbd or a shared drive, iscsi...) - the usage is step 2 - i think. It is okay to stay in a M/S State, because my goal is high availability. its nice if it goes to master/master state, but only nice.



The projected "miniplan" of consolidating my servers keeps care of taking local space (raid1 and redundant with drbd (protocol B) for ha-vms. All the "not ha needed" vms can take place on shared drives - shared drives will ever be a spof detail in my opinion.

What do you think - is proxmox ha ready at this time or should i build up my own ha systems - as i planned in the first step, before knowing proxmox.

Stefan!

I would souggest two things:
1. try pve on two test-machines which are kvm-able (btw. most of the amd-bords from asus support ecc-ram, so you get for cheap money an good system - ok not really server like (no ilom...), but stable and fast enough for normal things).

2. have a look at the beta of pve 2.0 - it's should arive this quarter.


Udo

 

[Hobbystern]

Good morning Udo,

i read your posting yesterday evening and after some sleep - i think i start from a new point of decision.

Everything sounds for me like its the best way to set up my own, well documented ha enviroment without proxmox. I am not new to the command line, so i think thats the easiest way to get what i want.

It makes no sense, to take proxmox - set up a cluster and do much work in the bash-background - only to have a web frontend and after 2 years i dont know where to start if the master or slave crashes, because everything is "done by hand". I hope you understand what i mean.

To your suggestions :

1. Try with i.e. AMD-V
Good idea - but i will reaching the point where i am now after some work. As i said is the 1st goal - high availability, not performance.
2. Beta 2.0
I will take a look on a test machine. My consolidation should start near by easter (not really much load on my servers, good time to switch). I hope the beta is out then, i will be patient.

Thanks Udo by now! WIsh you a sunny sunday!

EDIT: With a good cup of sleep i remember one thing. I told you - it is important to protect the sql server 2005 enterprise on a windows 2008 server vm.
I remember that 2005 Enterprise got a "cluster" connection, so that 2 sql server instances can take care of load balancing and failover - is this right i am not in need to build up a master/slave cluster, it can be a master/master cluster with two targets of disk space - the sql server then syncs the databases and i could use proxmox under this circumstances without using drbd. I hope i am right.

Stefan