[pedz]

This may seem like a thinly veiled attempt to advertise. And, if the system admins see it that way, then I apologize and I assume the thread will be removed.

I've developed SATA disk systems that can attach to pSeries (RS/6000) and AIX. The sizes range from 250G up to 20T per PCI slot. These are locally attached disks going at 3gb/s speeds -- no disk is faster. And many disks like fibre attached disks (SAN) are way slower.

When I approach IBM Partners to try and get them to help me sell them, they say "its too small". So, I'd like to know what they are talking about -- if someone can help me understand.

I *think* what a lot of sites do is they have all their machines connected via fibre to a hugh disk system and that is maybe 100T. It seems like the performance that would provide would be pretty bad compared to a locally attached disks like SCSI, SAS, or SATA. Maybe, a lot of sites don't need super speed but that would kinda surprise me.

It also seems like it would be harder to manage. The entire SAN network, the hubs, etc. It would seem expensive and a bit confusing. Compare that to each system having its own, locally attach, physically close by, disks.

But, no one seems interested for some reason. I'm trying to find out why.

Any help would be greatly appreciated.

[ross.mather]

The troubleis that you are going against the virtualisation trend. If the storage is locally attached it cannot be used for HACMP across boxes, it can't be used for Partition Mobility.

SANs are pretty easy to manage and many companies have dedicated teams for SAN storage and are moving away from local DASD entirely. Most pSeries systems I see now only have local DASD for VIO Servers.

The SAN performance is at its best when you put many heads to a single RAID array on the basis of many hands make light work. The speed they have in SAN environments is sufficient for most applications and SANs provide the flexibility to address performance issues.

I understand what you are suggesting - but you seem to be swimming against the tide of IT thought.

[pedz]

Originally Posted by ross.mather

The troubleis that you are going against the virtualisation trend. If the storage is locally attached it cannot be used for HACMP across boxes, it can't be used for Partition Mobility.

Sure it can. There are SATA enclosures that have dual host ports. "Local" and HACMP have long lived together. Plus, not that many people use HACMP.

I have SATA disks being used by VIO servers and client partitions as well. I can't think of any "virtualisation" that SATA can't do.

Originally Posted by ross.mather

SANs are pretty easy to manage and many companies have dedicated teams for SAN storage

That is a contradiction in the same sentence. If they are easy to manage, then why are "teams" needed to manage them?

Originally Posted by ross.mather

I understand what you are suggesting - but you seem to be swimming against the tide of IT thought.

This is clearly the case. Another trend is "RAID 5". RAID 5 made sense 10 years ago. Storage back then was expensive. Today, its cheap if you go with cheap solutions. Today, it is far cheaper, better performance, and easier to manage to just mirror everything. It is vastly simpler hardware and software.

I appreciate your reply -- but it adds to my frustration because the "logic" isn't there.

Thank you

[alexisl]

I guess the real question would be whether or not IBM supported it. Also, I think they might be right is suggesting that 20TB is too small, it might work well for a small business but then a small business isn't going to spend the money it would cost for IBM hardware/software.

Me personally, I would mind having something like that to put my backups on but only if it was supported by an IBM maintenance contract.

[pedz]

We are generally oppose to paid support. Our belief is that computers should be expected to work 100% of the time. If it doesn't, call us up and we'll fix it.

The whole concept of paid support, created because of the supremely bad software from Microsoft and others, goes against the original IBM policy of eternal support.

And, as far as being able to back up that claim, we are the guys who IBM vends out to to do the support.

But... no one has actually answered my question. What is the typical size of the storage a server has attached to it? Remember, the 20T is for a single HBA. Its trivial to put 5 HBAs into a server and have 100T on a single server.

[dig1tal]

I'm not sure what kind of answer you're looking for. The DS4000 series seems to be the sweet spot for mid-range storage. With the newer units using 270GB drives, you typically get 4.5T of raw storage out of 1 enclosure. Depending on the environment, some customers may allocate 1 enclosure to a specific LOB....for example, SAP non-production. Additionally, Zoning and LUN carving are pretty straight forward and not too complex for an AIX admin to handle.

Also, as Ross pointed out, the virtualization trend is increasing and storage virtualization is becoming a hot item.

[Hall Mil]

Not able to understand the things clearly, could you please be little precise?