Networker

Re: [Networker] Tape Compression defined

2002-12-04 14:05:01
Subject: Re: [Networker] Tape Compression defined
From: Terry Lemons <lemons_terry AT EMC DOT COM>
To: NETWORKER AT LISTMAIL.TEMPLE DOT EDU
Date: Wed, 4 Dec 2002 13:49:46 -0500
Hi Terry

Good question.  Nope, you should use the compressed data rates when
estimating total SCSI bandwidth, because the compressed data rates (the data
rates going INTO the tape drive) are what your system busses actually see
the data moving at.

For instance, let's take the case of sending two backup sessions to two DLT
7000 tape drives.  Each DLT7000 writes at 5 MB/s (uncompressed).  Both tape
drives connect to a single HVD SCSI bus that runs at a theoretical maximum
of 20 MB/s.  We'll assume that there are no limitations (choke points)
anywhere else in the environment.

-> one session is backing up .ZIP files (so this data can not be
compressed).  The maximum throughput into and through the tape drive will be
5 MB/s.  Because the data can not be compressed, the data moves into the
'front end' of the tape drive (where the compression engine is) at the same
rate that it moves to the 'back end' of the tape drive (where the 5 MB/s
tape movement mechanism is).
-> one session is backing up a variety of text files, which (in this
example) compress exactly 2:1.  Because the data can be compressed 2:1, the
data moves into the 'front end' of the tape drive at 10 MB/s, gets
compressed, and gets written through the 'back end' of the tape drive at 5
MB/s.

The limiting factor here is the 'back end' of the tape drive; it can never
go faster than 5 MB/s.  But, if you can compress the data stream, the
compressed data stream still gets WRITTEN to tape at up to 5 MB/s, but it
gets pulled INTO the tape drive front end at 10 MB/s.

So, these two backup streams generate (5 MB/s + 10 MB/s =) 15 MB/s.  Since
you're using a 20 MB/s bus, this is adequate for these two data streams.

Please let me know if I'm not explaining this well!
tl


-----Original Message-----
From: Lewis, Terry {Info~Palo Alto} [mailto:TERRY.LEWIS AT ROCHE DOT COM]
Sent: Wednesday, December 04, 2002 1:27 PM
To: Legato NetWorker discussion; lemons_terry AT emc DOT com
Subject: RE: [Networker] Tape Compression defined


Hey tl,

  Thanks for the explanation.  Does this mean that I
should only use the uncompressed throughput rates
and ignore the compressed rates when estimating
the total SCSI bandwidth required to support a given
number of tape drives?

tel

> -----Original Message-----
> From: Terry Lemons [SMTP:lemons_terry AT EMC DOT COM]
> Sent: Wednesday, December 04, 2002 7:37 AM
> To:   NETWORKER AT LISTMAIL.TEMPLE DOT EDU
> Subject:      Re: [Networker] Tape Compression defined
>
> Hi John
>
> Thoughts below.
> tl
>
> -----Original Message-----
> From: Ballinger, John M [mailto:john.ballinger AT PNL DOT GOV]
> Sent: Tuesday, November 26, 2002 2:34 PM
> To: NETWORKER AT LISTMAIL.TEMPLE DOT EDU
> Subject: [Networker] Tape Compression defined
>
>
> Tell me if I'm wrong but my understanding of tape compression is:
>
> Let's say I have a 50GB chunk of data available on a very fast
> disk/SCSI/PCI/etc.
> and let's say I can backup that 50GB chunk of data in 1 hour.
>
> Assume I'm using a DLT8000 tape drive which has a native streaming speed
of
> 6MB/s assuming no compression and a DLTIV tape in a DLT8000 drive with no
> compression on the data will fit 40GB of data on that tape.
>
> So that means my avg throughput is 50000MB/(1*60*60)sec or 13.88MB/s
> tl> Up to this point, I agree.
> And my compression ratio is 13.88/6  or 2.31:1
> tl> If you are assuming that no compression is being done, then your
> compression ratio is 1:1 here.
>
> And also during the backup on the average I will see the throughput
> indicated by the NetWorker Admin GUI as 13.88
> And furthermore I'm fitting 50GB of data(compressed) onto a tape that only
> holds 40GB in native(no compression)mode.
>
> tl> If you send an incompressible data stream to the DLT8000, and there
are
> no other bottlenecks, the tape drive will write at 6 MB/s.  But, if you
send
> a compressible data stream to the DLT8000, the tape drive will _appear_ to
> write at greater than 6 MB/s.  What actually happens is that the tape
drive
> subsystem can only write at up to 6 MB/s.  But the tape preprocessing
> subsystem (where the compression engine lives) can make the data smaller
> before it is sent to the tape drive subsystem.  So, you can actually have,
> say, 9 MB/s of uncompressed data going into the tape preprocessing
> subsystem, and 6 MB/s of compressed data coming out of the tape
> preprocessing subsystem and going into the tape drive subsystem.
>
> So, the effective tape write speed and the compressibility of the data are
> linked.
>
> Does this help?
>
> tl
>
> --
> Note: To sign off this list, send a "signoff" command via email
> to listserv AT listmail.temple DOT edu or visit the list's Web site at
> http://listmail.temple.edu/archives/networker.html where you can
> also view and post messages to the list.
> =*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=

--
Note: To sign off this list, send a "signoff" command via email
to listserv AT listmail.temple DOT edu or visit the list's Web site at
http://listmail.temple.edu/archives/networker.html where you can
also view and post messages to the list.
=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=

<Prev in Thread] Current Thread [Next in Thread>