ADSM-L

Re: D2D on AIX

2004-09-23 10:59:57
Subject: Re: D2D on AIX
From: TSM_User <tsm_user AT YAHOO DOT COM>
Date: Thu, 23 Sep 2004 07:05:51 -0700
Our server was collocated by node and we did have resource utilization set to 
10 with the  mount limit set to 10 so it used as many drives as it could.

Lets get back on track.  What I am saying is the seak on a tape costs time. 
Lots of time.  If you have a file server that has a high change rate then it 
will take longer to restore it a year after you start to back it up then the 
day after. With tape it will be much longer with disk not so much.  We have run 
a number of tests and we have seen time and time again that the fragmentation 
of a backup running for a year has a very small effect on disk but a large 
effect on tape.


Daniel Sparrman <Daniel.Sparrman AT EXIST DOT SE> wrote:
Hi

No, the restore test was not done with "new" data. The server had been
backing up data for several months, and utilized a total of 63 9840B
cartridges.However, mount time isnt an issue when having the possibility
of using multi-mount, multi-session restore. Its an even smaller issue
when having a tape library and tape drives that have a small time-to-data.

However, a large fileserver (+300GB) should never be non-collocated. This
will, as you say, spread the data over a non-acceptable number of tapes.
This however doesnt have to do with tape vs disk performance, rather TSM
design issues. A good design will always give you a good performance from
tape. The problem is, as servers are growing, no actions are taken to
withold good restore performance from the tape drives.

Designing the TSM system to have bad performance on either disk or tape
isnt that hard. The challenge is to design the TSM system with good
performance on both disk aswell as tape.

Mvh // Daniel Sparrman
-----------------------------------
Daniel Sparrman
Daniel Sparrman
Exist i Stockholm AB
Propellervdgen 6B
183 62 TDBY
Vdxel: 08 - 754 98 00
Mobil: 070 - 399 27 51



TSM_User
Sent by: "ADSM: Dist Stor Manager"
2004-09-22 19:16
Please respond to
"ADSM: Dist Stor Manager"


To
ADSM-L AT VM.MARIST DOT EDU
cc

Subject
Re: D2D on AIX






These numbers are from STK 9840B drives. I am not talking about a backup
and then a restore. I am talking about a daily backup for 8 months and
then a restore. File fragemenation dramatically effects your througput
over time. Sure we can spin 9840 drives to 100 GB/hr for large files. I
have even backed up a file server to 38 GB/hr (throgh 1 GB NIC's) with
millions of small files. But over time the speed is effected. It's being
unfair it is what it is.

For that restore test was it right after a the backup was done? What is
the change rate of the data. If it was after 8 months and you still got 40
GB/hr throughput then good deal. I haven't seen that.

Daniel Sparrman wrote:
Hi

Comparing these types of numbers are abit unfair. We have customers
running 9840 and LTO-2. They have alot higher throughput than 8-12GB/hour
over a GB nic.

For example, we have a customer running Netware. The TSM server is an AIX
server(pSeries 615) connected to a 3584-L32 library with 3 LTO-2 drives.
The Netware server has about 200GB of data. The AIX server has three
100Mbs nic, bundled togheter in an Etherchannel interface(theoretic speed
is 300Mbs or 30MB/s). The netware server is connected through 100Mbs
ethernet(single adapter). The server have a restore time of about 5= hours
which means we have an hourly throughput of almost 40GB/hour. Average
networkspeed is 11MB/s. The Netware server utilizes multi-session restore,
which means it can mount multiple volumes at once for restores.

We have another customer running a pSeries 650 clustrer. The cluster is
attached to a 3584-L32 library with 9 LTO-2 drives. The pSeries server is
equipped with an Etherchannel interface which consists of 2 GB nics.
During testing of a restore scenario on one of their Lotus Domino
servers(300GB of data), they reached about 50MB/s restoring directly from
tape. In this case, we didnt utilize multi-session restore, which meant
that the single LTO-2 drive could deliver 180GB/hour.

Today, the new tape technologies can easily outrun disks. To match LTO-2
drives against disks, you'll ned large, fiber-attached disk subsystems,
with no other load than the TSM server load. Internal SCSI-disks can never
outrun fiber-attached LTO-2 drives. The LTO-2 drive has a native speed of
35MB/s, compressed around 50-70MB/s depending on the type of data. They
also have dynamic speed, which means you dont get the back-hitch as long
as you keep writing data with at least 15MB/s. We've seen theese drives
push up to 90MB/s on database backups and restores. During the testing
phase of the implementation, we had up to 380MB/s from the disks(two
mirrored FAStT900 connected through 4 FC HBA:s with 34 15K 36.4GB fiber
disks per FAStT system) and almost 650MB/s from the drives(9 LTO-2 drives
connected through 4 FC HBA:s).

The speed of the drives is all about design. If you attach a large number
of drives to a single FC HBA, you'll easily get back-hitch. With the LTO-2
drives, a fair number of drives/adapter is around 3-4 / adapter.

Designing disk to match the tape drives is all about cost. S-ATA drives
can never outrun LTO-2 drives, at least not when it comes to large files
or database backups and restores. Designing FC disks to match the drives
will mean the cost is 10 times the cost of the tape drives.

This is all my opinion, and I'm sure that there are others out there that
dont agree.

Best Regards

Daniel Sparrman
-----------------------------------
Daniel Sparrman
Daniel Sparrman
Exist i Stockholm AB
Propellervdgen 6B
183 62 TDBY
Vdxel: 08 - 754 98 00
Mobil: 070 - 399 27 51



TSM_User
Sent by: "ADSM: Dist Stor Manager"
2004-09-22 04:27
Please respond to
"ADSM: Dist Stor Manager"


To
ADSM-L AT VM.MARIST DOT EDU
cc

Subject
Re: D2D on AIX






Good questions. Our real world example:We went from around 8 - 12 GB/hr
restore off of tape to over 40 GB/hr from the file device classes. Our
test was a file server with a little over 300 GB of data. The File server
and the TSM server both had 1 GB NIC's. Resource utilization was set to
10 in both cases. The data was fragemented on tape for a little over a
year for the first test. The data was fragmented over disk for nearly 8
months.

Steve Harris wrote: How does TSM access
the data on file volumes? Does it keep an offset of the start of every
file or aggregate?

If it does, then yes we could skip to the start of each file or aggregate.
If it does not, then we need to read through the volume to find the file
we are going to restore. Where we have a large number of concurrent
restores happening, this could cause performance issues on the array.

Now TSM has some smarts on later technology tape drives that have block
addressability and on-cartridge memory and can find a spot on the tape
quickly, but does this translate to file volumes?

Regards

Steve.

>>> lau AT VTCAT.CC.VT DOT EDU 22/09/2004 4:49:55 >>>
True. Seek time is tiny compared to tape mounts. I am just concerned that
the TSM db has to keep track of thousands of volume. How much will it
increase
the size of the db. Ours is already 90G at 70% utilized.

Eliza

>
> ==> In article <200409211619.i8LGJsbf018132 AT vtcat.cc.vt DOT edu>, Eliza Lau
writes:
>
> > What is the recommended volume size. I have seen someone mentioned 5G,
but
> > then the number of volumes will explode from about 800 (current # of
3590
> > primary tapes) to thousands.
>
> Consider, this doesn't really cost you much. Seek time in a directory of
> thousands of files is still tiny compared to tape behavior.
>
> I probably wouldn't go as low as 5G, but 10G (much less than the average
size
> of my 3590 vols) seems pretty reasonable to me. 20G is getting big, from
my
> perspective.
>
>
>
> > How about keeping the staging space so clients backup to staging then
> > migrate to FILE volumes. Then every volume will be filled up.
>
>
> I like this, too.
>
>
> - Allen S. Rout
>



***********************************************************************************
This email, including any attachments sent with it, is confidential and
for the sole use of the intended recipient(s). This confidentiality is not
waived or lost, if you receive it and you are not the intended
recipient(s), or if it is transmitted/received in error.

Any unauthorised use, alteration, disclosure, distribution or review of
this email is prohibited. It may be subject to a statutory duty of
confidentiality if it relates to health service matters.

If you are not the intended recipient(s), or if you have received this
email in error, you are asked to immediately notify the sender by
telephone or by return email. You should also delete this email and
destroy any hard copies produced.
***********************************************************************************



__________________________________________________
Do You Yahoo!?
Tired of spam? Yahoo! Mail has the best spam protection around
http://mail.yahoo.com

__________________________________________________
Do You Yahoo!?
Tired of spam? Yahoo! Mail has the best spam protection around
http://mail.yahoo.com


---------------------------------
Do you Yahoo!?
Do you Yahoo!?
vote.yahoo.com - Register online to vote today!
<Prev in Thread] Current Thread [Next in Thread>