New iOS mobile app monitors server admin
How about an MPE hobbyist's advocate?

How LTO Tape Support Won't Matter, Soon

LTOBetterA few weeks ago an InfoWorld article told the IT community that the storage in the cloud was the final nail in backup tape's coffin. Our intrepid author Brian Edminster took a close look at what the Amazon Glacier cloud could do for the HP 3000 user. But it's almost as important to listen to what he's got to say about support of the latest LTO tape devices.

They won't make you need to migrate, though, if you virtualize the 3000 iron.

It's just another example of how an emulator removes the risk of staying on an environment. A virtualized server isn't going to be tied to interfaces from 10-year-old systems, or IO designs first crafted in the previous century.

This used to be a big deal in HP's engineering plans. One of the primary advantages to creating PA-RISC architecture was supposed to be peripheral support. HP figured to be writing and maintaining fewer device drivers if its enterprise servers shared an architecture. PA-RISC just led HP away from the HP-IB interface, something Hewlett-Packard created for instruments, not computers. But in practice, the operating systems still needed specialized engineering to pass data quickly between server and peripheral.

These late-gen LTO-5 tape drives are the kind of peripherals which HP supported more slowly, if at all, during the final decade of lab work on MPE. The first LTO with an HP badge, Ultrium, ran half as fast (160 mb/sec) as the same unit hooked to HP-UX -- because its mandatory MPE interface was engineered for half the bandwidth of the more updated Unix-based servers. HP never made up the difference in speed, and that shortfall arrived right out of the gate with LTO-1. LTO-5 was the state of the art in 2010, two years after HP closed the MPE labs.

Aging backup devices can pose a serious reason to consider a migration off the 3000 iron, if you're bound to an HP-badged box. The media gets harder to buy. The devices become a special case for IT to support -- although there are some crack independent companies who'll service 3000 sites regardless of what backup drives are on the job.

Emulation -- the virtualization engine of the 3000's hardware -- changes all that. Edminster said if a VM supports a device, then the aging artifact of peripheral interface simply goes away. Supporting tape devices was a milestone which the Stromasys emulator crossed early in 2012. "I think that the question of should MPE/iX have support for LTO-5 is largely a red herring," Edminster says.

In a solid virtualization design, whatever device the hosting hardware supports (in the current emulator's release, that's any Intel i7 Core system) is the only thing that matters. And if the cloud replaces tape, fine. But you won't need to rely upon cloud storage just because HP stopped engineering MPE's IO a decade ago. Edminster explains.

LTO-5 is largely a red herring. Why? Because it doesn't matter if the 3000 support it or not. Instead, does the hosting VM support it? My guess is that the only instances of MPE/iX which will survive, in the longest term, might be those which run under the Charon HPA/3000 VM. Since the hosting VM manages the disk images and their backup, it'll all be transparent to MPE/iX as to what kind of medium is being used. That's true if the backup occurs via a 'virtual' tape drive, or even that it's being backed up at all (a backup of disk-image, done by the hosting VM).

LTO was not a project in HP's labs that got extensive 3000 testing victories -- that's to say, a wide scope of software running against it which passed the MPE/iX speed tests. Jim Hawkins, the IO device expert in that lab, says the tests failed to deliver adequate small-file transfers using HP's own backup software, TurboStore.

When Herb Statham of Cerro Wire asked if he could use a LTO-1 Tape Drive on an A-Class 500 HP 3000 with Turbo/Store iX, because his was back was backup exceeding 100GB, Hawkins had one word of advice: Don't.

Hawkins referred to a page of a 2004 HP Communicator, a tech document written in support of the PowerPatch 2 release of MPE/iX 7.5. That's just about the last Communicator that HP produced about 3000 techology. The warning on page 26 sets expectations pretty low for Ultrium LTO. The 215 and 230 models were the state of HP's art in 2004.

Physical connections are to be made only to LVD-SCSI Host Bus Adaptors. LVD-SCSI terminators must be used for devices to function at rated speeds. HP recommends only one Ultrium Tape device per SCSI bus for maximum performance. No more than two Ultrium Tape devices per SCSI bus will be supported. An Ultrium device must never share a SCSI bus with any other SCSI peripheral type.

There's also the matter that there was little support for using MPE/iX to diagnose Ultrium problems .

Most diagnostic support for Ultrium drives comes from HP Storage Works Library and Tape Tools (a.k.a. LTT). LTT does not run on MPE/iX; therefore in some diagnostic scenarios the Ultrium may have to be removed from the HP e3000 and connected to a host running LTT.

So here comes HP LTO-1 technology that was too advanced to work with HP's own backup software. The indie software tools HiBack and Backup+/iX were the only backup apps certified for Ultrium. Not TurboStore, "for the reason of poor performance, especially for small files," Hawkins told me.

But don't interpret that "don't" too literally. It's not that LTO devices of that era are unusable with 3000s. Not at all. Consultant Craig Lalley of EchoTech reports that there's N-Class servers in his client base using LTO-1. Hawkins said the tests against TurboStore didn't pan out, at least for the little things. Like files.

Basically performance may be very much less that "Native" device speed, even slower than DDS-4 in some cases, due to a combination of TSTORE and TapeDM limitations. In fact we'd already seen a bit of a drop-off in TSTORE performance with DLT80/8000.

Thinking about it again, I suspect that customers with a set of very large files would probably do okay, especially if you have the space a store-to-disk backup and then a store of those files to Ttpe probably would be okay.

It's the small files that will get you hung up using LTO-1. Hawkins even shared his lab notes from those tests, for the customer who's tech-savvy enough to want details on the failed proof of concept.

It is apparent from the TSTTOOL results that the larger the blocks being written, the faster the Ultrium device will run.  Also the fewer the number of file marks, particularly on smaller block sizes, the faster Ultrium will run. Although the numbers achieved by TSTTOOL are not realistic compared to STORE since no disk IO are required to deliver the data to the tape. It still demonstrated the potential for improvement. Even STORE shows some improvement depending on whether the MAXTAPEBUF option is used. I would recommend that the MAXTAPEBUF be increased to 64, or possibly 128K.

Secondly, the combination of file marks and small block sizes can be devastating. The STORE test with the statistics option clearly shows that the Ultrium and the DLT80 are greatly effected by the storing many small files due to file mark usage between file. I am assuming that the use of file marks is tied to using the SPACE command to moving around on the tape currently.  

So, I would recommend that the file marks be reduced or eliminated through the use of alternate positioning commands; i.e., READ POSITION and LOCATE which allows the device to move quickly to any point on the tape.  If  READ POSITION and LOCATE are considered the tape DMs will require updating as well.

Something not addressed in this investigation that may (or may not) need to be checked: to ensure that the 200Gb cartridge, capable of handling many more files than previous tape devices, does not have any issues handling a potentially large directory size for a tape or combinations of tapes.

If you don't have a copy of that 7.5 PP2 Communicator handy, you can download it from us here. The link will probably outlast LTOs 1-4, and even LTO-5.