Questions

Volker Wiegand wiegand at suse.de
Tue Nov 2 09:16:53 MST 1999


On Tue, 2 Nov 1999, Giampaolo Tomassoni wrote:

> Hi all,
> 
> I have a couple of questions and, possibly, suggestions to submit to this list.
> 
> I work as an IT consultant among many small companies, where I see there is an 
> increasing awareness about data safety and system availability, to the extent 
> that even very small companies show an interest in HA solutions. This is 
> probably because companies of every level already faced at least once the 
> expenses and troubles related to a system failure or disk corruption.
> 
Exactly.

> In this sense, I started looking for inexpensive HA solutions for business 
> software, which could eventually well fit to very small businesses (less than 
> 500K$/year gross income).
> 
> This basically meant some type of HA implemented in top of Linux to me, since 
> any other choice (IBM HACMP and Microsoft Cluster) imposes costs which may get 
> the product unattractive to many low-end buyer.
> 
> Recently, I reached the Linux/HA site, which challenged me many questions.
> 
Not only you :-)

> It seems that the hardware Linux/HA architecture is somehow a replica of the 
> IBM/HACMP (from which the project seems to fetch most of the basic ideas). The 
> basic architecture doubles the machines, the network interfaces, allows SCSI 
> and suggests SSA for the attachment of disk devices, doubling the SCSI/SSA 
> interfaces and disks, etc. etc.
> 
Well, the Linux-HA project took many ideas and terminology from the Linux
HA HOWTO which was originally authored by Harald Milz, who was then
working for ... IBM.  Of course Harald has transferred a lot of the good
ideas from HACMP to the HOWTO, and many people still think HACMP is not
too bad.

At the time of the original writing SSA was a promising way for HA, but
today he would have probably written more on FC.

> Isn't the final system out of the extents of Linux? For few K$ more, wouldn't 
> you, the boss, buy an IBM/HACMP system instead? Where is the advantage of using 
> Linux?
> 
Have you ever purchased and maintained an HACMP system?  If I has these
"few K$ more", life would be easier ...

> Keeping in mind the business market, I believe that the OpenSourceness of Linux 
> over AIX is less than an issue.
> 
> Now, I have the feeling that we can cut the costs down by:
>  - Not using SCSI nor SSA, whose interface cards
>    and disks are too expensive.
> 
That depends on your requirements of share level (your approach works
quite ok for shared nothing only) and quality.  Try to sell EIDE drives to
someone with mission critical needs !!!

>  - Not using external storage units, which require at
>    least a case, a double power supply, an interface
>    (again SCSI or SSA) and, to avoid to became a spof,
>    must be at least two. Not even to mention external
>    raid units!
> 
Hmmm, the costs are not that high, me thinks.

>  - Focusing development efforts more on less on the
>    typical Linux hardware, which is around EIDE storage.
> 
Why narrow?  The "typical Linux hardware" is more flexible than you might
think.

> I believe that a typical redundant HA system running Linux would be made of two 
> machines, each configured more or less this way:
>  - A network card toward the clients network.
>    IP aliasing can be used on a takeover.
>  - A fast network card (or other fast mean) to
>    interconnect the servers.
>  - At least two fast EIDE disks, mirrored through software
>    RAID level 1.
>  - An RS232C interface to replicate the heartbeat and status
>    interconnection.
>  - Linux, of course. With HA software on top, too.
>  - Some kind on networked mirroring (!).
>  - GFS, or other kind of networked fs, to coordinate fs
>    accesses.
> 
Hmmm, I'm not sure if GFS is already ready for production use.

Besides, your suggested configuration is what the Linux-HA project
currently uses and runs on.  Please don't take the suggestions of the
Linux HA HOWTO in its historical state for the Linux-HA project.

I have recently taken over responsibility for the HOWTO and am currently
reworking it.

> I see no spofs and many advantages (among which the low cost, but also the 
> scalability) on this configuration, am I right?
> 
Yes.

> Assumed I am, what I miss to do something like this is the network mirroring 
> stuff. I have very few ideas on how to realize it, and I don't know if they are 
> really feasible nor how much time would be needed to implement it. Probably I 
> am missing something from the Linux/HA site, but I see nothing about it. Maybe 
> there is already something somewhere, somehow close to what would fit these 
> architectural requirements, but I cannot find it. Is there a standard (or 
> sources, better) for networked raid? Is there any provisioning to implement, 
> support or integrate it in Linux/HA?
> 
It's already there, almost.  Look for NBD (Networked Block Device) on the
mailing list.  Think it was called something like "poor man's data
replication" in the mail titles.

> What are your feelings about this solution?
> 
Warm :-)

> A final question, for which you probably know the answer. What is GFS about? 
> Can it handle backup routes for data sources, or rather is just a method to 
> 'stripe' toghether disk units? The GFS paper seems to miss a 'goals' section...
> 
I think their main goal is to provide storage independent of machines.

> Thanks, and sorry for being a bit long and confusional in my message...
> 
No, you were clear.  Long messages have a chance for continued
communications :-)

> Giampaolo Tomassoni	Information Systems Consultant
> 
Volker

--
Volker Wiegand               Phone: +49 (0) 6196 / 50951-24
SuSE Rhein/Main AG             Fax: +49 (0) 6196 / 40 96 07
Mergenthalerallee 45-47     Mobile: +49 (0) 179 / 292 66 76
D-65760 Eschborn            E-Mail:  Volker.Wiegand at suse.de
++ Only users lose drugs. Or was it the other way round? ++




More information about the Linux-HA mailing list