DLT or LTO or AIT for new tape drive?

[Rob Turk]

That's dishonest. Others do as reported here by Steve Cousins.
Unless they mean the only "consumer tape drive." In which case it's
misleading but correct AFAIK.

VXA is one of the few that does dynamic speed adjustment, and the only
technology that does this over a broad speed range.

Linear tape drives can slow down to some extent but they need to adjust the
write frequencies that go through the heads to ensure the same bits per inch
are on the tape. Tape drive heads are optimised for specific frequencies,
slowing down too far would make it very hard to write a consistent signal on
tape. LTO and SDLT can do it, but not in order of magnitudes. They can't
slow the tape down to, say, 1/10th of the speed.

Helical scan can slow down the speed of tape in steps quite easily. You can
write half speed with the same bits/second simply by moving the tape at half
the speed and skipping one revolution of the write head. The next step is
1/3rd speed where the head skips 2 revolutions before writing the next
track. You can keep slowing down easily to 1/10th or further without
changing the write frequency signals. The issue is that the track angle will
be diffent for each speed step. All helical-scan drives depend on servo
signals to follow the track on read. When the track angle changes, the read
heads can't follow the track. Therefor this doesn't work on drives that
depend on servo signals.

Enter VXA. It's packet technology does not rely on servo signals to read
back the tape. It scans the entire surface and picks up every single packet
fragment, regardless the angle it was written with. It re-assembles the data
in buffer memory and delivers it back to the host. VXA is therefor the only
technology that can slow down substantially to match the host data speed.

Rob

 

[Curious George]

Curious George,

I must say. You seem to be very knowledgeable about tape drives.

Hmm. Even though I use SDLT 220's but didn't know/notice they could
do speed compensation? Thanks for the vote of confidence. I'm still
red in the face.

Are
you a storage administrator for a large SAN or do you work for a drive
company?

Don't worry, I won't make a dime whether you choose VXA, AIT, (S)AIT,
DLT, DDS, etc. I have no professional investment in any product.

Anyway, the comment above. Are you implying that VXA is not in the
same league as say LTO or DLT? Why? Quality? Performance?
Durability of the media?

Sort of. Probably more due to price & marketing than
quality/durability - at least to the extent that I've never used VXA &
have only heard good things about it. Of course these rave reviews
thend to come from "power user workstations" & not anything extreme.
The real track record in the enterprise is DLT with LTO/Ultrium
proving itself quite well. (S)AIT is a smaller market share but also
viable

Capacity, performance, tape price, and library configurations are
other considerations which nocks VXA slightly below those others. The
1x10 autoloader seems really nice for a small business or departmental
server. It seems to me the exebyte marketing is focusing at that
market share & below. You can get larger (albeit older) libraries
with VXA, but it is not common or available among many brands.

The tapes are also expensive per MB and small & slow compared to some
other technologies. This a significant issue with larger backup jobs
& with even basic retention that makes other more expensive drives
cheaper in the long run.

The two separate ideas are (1) compression and (2) variable tape
speed? Right?

Correct. One is "synthetic," artificially inflated data rates that
fluctuate basically per file or job's compressibility. The other is a
"real" change in tape speed; a compensation to match the host system's
throughput.

That sounds too complicated to me. If a drive isn't capable of
hardware compression your backup program can likely do it's own
compression if you tell it too. Whether the compression happens on
tape firmware or host system, compression rates will be based on the
algorithm & files. If the host can't keep up either the tape has to
slow down (if it can) or there is a buffer underrun and shoeshine. I
don't imagine a tape engineer also putting in code that if the tape
can't slow up enough compression rates are reduced on the fly. It's
possible but I don't think anyone bothers. Esp as this can involve
some coordination with backup software & tape compensation is often
not that extreme.

Got it. Sounds like you are pointing me to DLT and LTO, but not AIT,
and you're suggesting that VXA isn't as good as these two.

Many have found VXA to be reliable. I'm not trying to steer you away
from it. I just got into a discussion about other tapes than can do
variable speed compensation because you said that's important to you
and that Exebyte claimed they are the only ones who do it.

(S)AIT should also work well even though it hasn't been the industry
standard enterprise tape like DLT. Like VXA, AIT is a helical-scan
technology (unlike DLT & LTO). Unlike VXA I don't believe it does
speed compensation (but I could be wrong again)

Obviously I will need to be concerned about price. Now, for my
"consumer" application backing up a TB+ of digital photo data, where I
won't be running the drive non-stop, would VXA be OK,

The problem is a TB+ of data. That already puts it beyond the scope
of a normal "consumer application" so "consumer grade" gear & prices
don't apply. Of course I'm not saying VXA is a flat out "consumer
grade technology". It isn't Travan. Not by a long shot.


Lets examine the situation a little:

If your digital photos aren't raw they are likely already pretty
compressed so you need to look at native tape capacity and tape count.
Don't make a purchase decision based on compressed capacity & rate
claims. You should never expect to get full or near 2:1 compression
for anywhere near the entirety of a normal backup job.

So a complete backup & verify will be a TON of tape changes (almost 30
over the course of a couple of days for VXA2- with the drive being
idle quite a bit waiting for you) and therefore totally impractical to
do manually. The VXA2 1/10 autoloader is insufficient for your
current backup job. You should always allow for projected growth
also. I don't think the VXA-320 is available in an autoloader yet
(worth checking though). If it is its new so prices would be high.

I'd also suggest that it is irresponsible to have anything less than 3
separate full backups on 3 different media sets spread out over time
(& hopefully location). This is really _bare_minimum_ & it's also a
lot of tapes (~39-45 for 1TB on VXA2) There's so much DLT floating
around that new tapes can be bought cheap. LTO was made from the
beginning to be a good $$/MB or tape with equal or better performance
& reliability as the competition.

Just keep in mind that a 1/2 TB can take about a day (around 20 hours)
to run a full backup and verify on a SDLT 220 autoloader. VXA2 is 1/2
the speed and will use more tapes. Basically for VXA you're going to
need something like an exebyte 430 Library with VXA2 (30 carts up to 4
drives) and a _very_large_backup_&_restore window. Even RAIT might
make some sense because of backup/restore windows (ON second thought
probably not).

With 1 TB I actually would really steer you towards an LTO2 or AIT-4
or higher autoloader with a barcode reader. (The SDLT320/DLTV4 may
make things a bit tight and you may find the SDLT600 too pricy.) I'd
also ask you to project you data growth & also determine how you want
to handle incrementals/differetials (# slots, etc). Since price is a
concern you're basically forced to look at the used/refurb market. I
wouldn't turn my nose up at open box, used or refurbs or believe tape
is "too expensive" for "consumer" use. It's expensive, but it's also
the basic cost of computing. 1TB is not that easy or fast or safe to
backup properly shuffling HDD's around. Unfortunately there's really
nothing else. The more complex & handholding you have to do for a
backup, the less likely you will actually do it down the line. If it
isn't simple, fast, easy & comprehensive (& hopefully somewhat or
totally automatic) you'll probably end up with no backup at all when
you need it.

 

[Lady Margaret Thatcher]

Lots of stuff snipped.


That's what I wrote last time. After reading your last posting, I
would "double that" statement. You are truly awesome. If you are
blushing like crazy well then you probably don't realize how awesome

Don't worry, I won't make a dime whether you choose VXA, AIT, (S)AIT,
DLT, DDS, etc. I have no professional investment in any product.

Yeah, but you must be a SAN administrator in some enterprise
somewhere. With certifications from people like BMC or Network
Appliance and Veritas.

Capacity, performance, tape price, and library configurations are
other considerations which nocks VXA slightly below those others. The
1x10 autoloader seems really nice for a small business or departmental
server. It seems to me the exebyte marketing is focusing at that
market share & below. You can get larger (albeit older) libraries
with VXA, but it is not common or available among many brands.

Good point.

The tapes are also expensive per MB and small & slow compared to some
other technologies. This a significant issue with larger backup jobs
& with even basic retention that makes other more expensive drives
cheaper in the long run.

Now price IS an issue, especially after I start accumulating enough
tapes to do the triple-backup you suggest below, and for several
generations. E.g. Do a full backup every six months, then incremental
backups. And retain 1-2 previous generations of backup. By the way,
I use Dantz Retrospect, which I like for their "incremental plus"
approach.

Correct. One is "synthetic," artificially inflated data rates that
fluctuate basically per file or job's compressibility. The other is a
"real" change in tape speed; a compensation to match the host system's
throughput.

Got it. Just wanted to confirm my understanding.

That sounds too complicated to me. If a drive isn't capable of
hardware compression your backup program can likely do it's own
compression if you tell it too. Whether the compression happens on
tape firmware or host system, compression rates will be based on the
algorithm & files. If the host can't keep up either the tape has to
slow down (if it can) or there is a buffer underrun and shoeshine. I
don't imagine a tape engineer also putting in code that if the tape
can't slow up enough compression rates are reduced on the fly. It's

actually I can't either. Just wanted to ask.

possible but I don't think anyone bothers. Esp as this can involve
some coordination with backup software & tape compensation is often
not that extreme.



Many have found VXA to be reliable. I'm not trying to steer you away
from it. I just got into a discussion about other tapes than can do
variable speed compensation because you said that's important to you
and that Exebyte claimed they are the only ones who do it.

(S)AIT should also work well even though it hasn't been the industry
standard enterprise tape like DLT. Like VXA, AIT is a helical-scan
technology (unlike DLT & LTO). Unlike VXA I don't believe it does
speed compensation (but I could be wrong again)

OK. I'll rule out AIT. That still leaves three different tape
technologies that I need to compare, including tape costs.

The problem is a TB+ of data. That already puts it beyond the scope
of a normal "consumer application" so "consumer grade" gear & prices
don't apply. Of course I'm not saying VXA is a flat out "consumer
grade technology". It isn't Travan. Not by a long shot.

I won't have a full TB+ of data on "day 1." More likely, over Years
1, 2, and maybe 3. But I do intend to digitize my entire "analog"
photo collection and switch over to digital in about 12-18 months.

Lets examine the situation a little:

If your digital photos aren't raw they are likely already pretty

Actually I will be shooting RAW, since that is the "digital negative".
But every negative needs to be turned into one or more "positives" so
I could easily have 3, 4, or maybe even five files for a given image.

compressed so you need to look at native tape capacity and tape count.

Yes. However, I will probably be working with NEF files, which are
compressed Nikon RAW files, so your point is still valid.

Don't make a purchase decision based on compressed capacity & rate
claims. You should never expect to get full or near 2:1 compression
for anywhere near the entirety of a normal backup job.

Even today, I never get full 2:1 compression. I think that's one of
the "Ten Big Lies," along with, "The check is in the mail," and
similar statements.

So a complete backup & verify will be a TON of tape changes (almost 30
over the course of a couple of days for VXA2- with the drive being
idle quite a bit waiting for you) and therefore totally impractical to
do manually. The VXA2 1/10 autoloader is insufficient for your

Agreed. What got me started on this whole idea of upgrading is that
right now, each full backup of my home LAN needs three tape changes,
and that's already too many. I'm not sure I want to get an
autoloader, but ... I may do that after I figure everything out.

current backup job. You should always allow for projected growth
also. I don't think the VXA-320 is available in an autoloader yet
(worth checking though). If it is its new so prices would be high.

I'd also suggest that it is irresponsible to have anything less than 3
separate full backups on 3 different media sets spread out over time
(& hopefully location). This is really _bare_minimum_ & it's also a

Agreed. Even today, I keep a "critical offsite backup" in my office.
If I was paranoid, I would ship the tapes out of state, or at least
out of the same weather area. (c.f. Katrina, Rita, Wilma?)

lot of tapes (~39-45 for 1TB on VXA2) There's so much DLT floating
around that new tapes can be bought cheap. LTO was made from the
beginning to be a good $$/MB or tape with equal or better performance
& reliability as the competition.

Well, that has me thinking that maybe I should focus on DLT.

Just keep in mind that a 1/2 TB can take about a day (around 20 hours)
to run a full backup and verify on a SDLT 220 autoloader. VXA2 is 1/2
the speed and will use more tapes. Basically for VXA you're going to

The half-speed isn't a problem for me, since I'm concerned about
buffer underruns. But the number of media changes is an issue.

need something like an exebyte 430 Library with VXA2 (30 carts up to 4
drives) and a _very_large_backup_&_restore window. Even RAIT might
make some sense because of backup/restore windows (ON second thought
probably not).

This is getting to sound pretty pricey.

With 1 TB I actually would really steer you towards an LTO2 or AIT-4
or higher autoloader with a barcode reader. (The SDLT320/DLTV4 may
make things a bit tight and you may find the SDLT600 too pricy.) I'd
also ask you to project you data growth & also determine how you want
to handle incrementals/differetials (# slots, etc). Since price is a
concern you're basically forced to look at the used/refurb market. I

Agreed. EBay is my real friend here. That is how I can afford to
have four SCSI Ultra 320 10K drives on my system, with an LSI
"midrange" host adapter with built-in RAID. I paid less than WD
Raptor prices for these drives and they are FAST!

wouldn't turn my nose up at open box, used or refurbs or believe tape
is "too expensive" for "consumer" use. It's expensive, but it's also

Not at all. In fact, I would turn up my nose only at paying
retail/new.

the basic cost of computing. 1TB is not that easy or fast or safe to
backup properly shuffling HDD's around. Unfortunately there's really
nothing else. The more complex & handholding you have to do for a
backup, the less likely you will actually do it down the line. If it

So true.

isn't simple, fast, easy & comprehensive (& hopefully somewhat or
totally automatic) you'll probably end up with no backup at all when
you need it.

Well, for me, this is just part of the cost of the "digital darkroom,"
along with the really nice Epson pigment-ink printer, the film
scanner, the calibration system to match color between camera,
monitor, scanner and printer, etc. Plus some software to manage this
quantity of images. And Photoshop, of course.

I'm going to go way, way out on a limb here and say that you're a
usenet old-timer. This is the way the usenet was 15-20 years ago when
everything was uucp "bang-path" and was routed through inhp4 or
ucbvax.

--Lady Margaret Thatcher--

 

[Lady Margaret Thatcher]

VXA is one of the few that does dynamic speed adjustment, and the only
technology that does this over a broad speed range.

Linear tape drives can slow down to some extent but they need to adjust the
write frequencies that go through the heads to ensure the same bits per inch
are on the tape. Tape drive heads are optimised for specific frequencies,
slowing down too far would make it very hard to write a consistent signal on
tape. LTO and SDLT can do it, but not in order of magnitudes. They can't
slow the tape down to, say, 1/10th of the speed.

Helical scan can slow down the speed of tape in steps quite easily. You can
write half speed with the same bits/second simply by moving the tape at half
the speed and skipping one revolution of the write head. The next step is
1/3rd speed where the head skips 2 revolutions before writing the next
track. You can keep slowing down easily to 1/10th or further without
changing the write frequency signals. The issue is that the track angle will
be diffent for each speed step. All helical-scan drives depend on servo
signals to follow the track on read. When the track angle changes, the read
heads can't follow the track. Therefor this doesn't work on drives that
depend on servo signals.

Enter VXA. It's packet technology does not rely on servo signals to read
back the tape. It scans the entire surface and picks up every single packet
fragment, regardless the angle it was written with. It re-assembles the data
in buffer memory and delivers it back to the host. VXA is therefor the only
technology that can slow down substantially to match the host data speed.

Rob

Rob,

Thanks. I'm not sure how much of a "slowdown" effect I would really
need. If I did a disk-to-disk copy of some large files and measured
the throughput, could I assume that I would get the same throughput
into a backup tape drive?

--Thatcher--

 

[Curious George]

Thanks for the gushing praise. Old farts like me love to be flattered
by former heads of state.

I thought you were starting with 1TB+ data on day one. That may
change things significantly - depending on how that project panns out.
a few more words:

I want to stress again calculating storage reqs based on native
capacity and minimizing media changes. See the compression as more of
a bonus. "The checks in the mail", etc.

Unless you are going with something very high capacity like SAIT, etc
or are sure this project won't happen, an autoloader might be more
realistic than you think esp if you are a patient shopper. Last year
I picked up another L200 SDLT220 new/open box complete retail parts
around $1300 delivered. No one seemed to want it because it was HVD
(I guess). It really wasn't all that much more than a new VXA2 (at
the time ~$1000). New sealed SDLT1 tapes can be auctioned off $20-25
(when lucky) or new in stores a little over $30 apiece. Not the
typical "consumer" investment but not as imposing as one might think.
(unfortunately, though, I don't think it's a great fit for 1TB+)

Listening to a recent backup the term "dual speed recording" makes
more sense to me. It sounds like there is a "fast" & "slow" speed.
This observation ostensibly corroborates Rob Turks comments and makes
more sense to me why it wouldn't be advertized very heavily, esp not
as "variable speed compensation" & the like.

Triple backup is something to consider as a starting point. It all
depends on how quickly the data grows. You may be happy initially
with monthly's that turn into monthly & quarterly & then at a busy
time need to create dailies. Then you might want to free
up/consolidate pools at year end. You couldn't be quite so flexible
for a business network but since it's just you, you can easy adjust
things as you go & controll/limit the use of media while maximizing
ease of recovery without too much effort. At least with robotic tape
you're only thinking of jobs & pools and not the details of fitting
what onto which media.

Take your time. Plan and then rework the plan. It'll take some
planning to get a good idea of the real cost of this project &
retaining the images this way as opposed to other options. You're off
to a good start. Just don't forget to encrypt the backups, esp
offsite as well as all the other basic tape best practices.