Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

Counter:

Retrieval (recovery) from disk is easier and faster - ability to randomly read and write. Tapes are sequential - to get to a data which is only at the end, one has to go through all the tapes that is there before that.



The maximum tape seek time is around a couple of minutes. The average seek time is about half of that, i.e. around 1 minute (these example values are valid for LTO-7).

These seek times are seldom a problem. If one would keep on HDDs as much data as it is normally kept on tapes, i.e. at least a few hundred terabytes, one would have the same slow seek times as someone would have to plug and unplug external HDDs, unless the storage would belong to a very large company that could afford the high costs of keeping online many HDDs.

As an individual user, I keep in my computer, on an SSD, an index with the content of all tapes. When I need some file, I get it in at most 5 minutes, which includes starting the tape drive, going to a cabinet and taking the tape from a shelf, inserting it in the drive and waiting for the seek and copy operations.

This is perfectly acceptable, because I do not need files from the tapes every day. Storing the data on external HDDs would not reduce the time wasted with manual operations, it would be much less reliable and it would be much more expensive for great amounts of data.

The sequential transfer speed of tapes is greater than that of HDDs. Therefore, if after a seek you copy large files, e.g. a movie in BluRay format, the whole seek + copy operation takes less than from a HDD.


That detail is abstracted away from the user though, tapes north of LTO 5 (which is pretty old) allow addressing tapes like a normal file system, and if you are doing backup things, the contents of the tape are held separately from the data itself.

As an off-line archival backup medium, access times really are not a concern in the real world.


Tapes are not strictly sequential, they are also a 2D medium, admittedly with one sequential dimension:

https://en.wikipedia.org/wiki/Linear_Tape-Open#Physical_stru...

https://en.wikipedia.org/wiki/Linear_Tape-Open#Positioning_t...


What is the value of that? Is that a common use case for those buying exabytes of tape?


yes, for random IO tapes suck.

but, when you are recovering files, its rarely a random io type deal.

Even then, you generally just dump to the nearline/tape cache and fiddle with the data there.

A decent 25 drive tape library will easily saturate a 100gig network link, and its perfectly possible to add more drives to get more IO.

Tapes as part of tiered storage is something that is really powerful. Yes, flash is cheap, but not cheap enough to be used as long term backup (ie legal hold or general archive.)

Keep your expensive fast storage near your clients, the cheaper less fast, but more voluminous a stage away, and then dump out to tape after n weeks with no access.


If you are using tapes, you aren't going to use it for a use case where you need to recover specific parts of the backup.

Tapes are for archival storage, not for 'accidentally deleted a single file and need it back' type backups.


That’s just not correct. Tapes can be used for either purpose, and with a decent backup software managing them, restoring specific parts of the backup is easy—you just need to wait for it to seek the tape.


Indeed. The only time I have ever seen anyone restore from tape in case of real need was to recover a tiny fraction (< .02%) of files that software had mistakenly deleted. https://gmail.googleblog.com/2011/02/gmail-back-soon-for-eve...


This is incorrect. Last place I worked at with tape, the poor tape guy was constantly going back for some deleted a single file and needed it back situation.

Once, we had to set up a hollow Sharepoint and restore a Sharepoint backup just to get someone's deleted file.

Now, do I think this is a good idea? No. Frankly, people who cause these kinds of things need to see "This cost X hours to recover, stop doing that" as feedback.


The disk also has to move its head to the location of the data. It is faster, but still sequential.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: