Hi Daniel,
This things we already have try and don't work.
That's why we are moving over to multiple servers and I have explain this for
them.
But if the customer wanna try TSM Server on VM, then why not. It is always a
fun test to do.
Thanks for everyonce respond and feedback.
I will do some testing during Q1 and Q2 and if anyone are instested of the
resault you are welcome to drop me an email.
Cheers
Christian
Sent from my Sony Ericsson XPERIA™ X1.
----- Ursprungligt meddelande -----
Från: Daniel Sparrman <daniel.sparrman AT EXIST DOT SE>
Skickat: den 2 februari 2009 14:59
Till: ADSM-L AT VM.MARIST DOT EDU <ADSM-L AT VM.MARIST DOT EDU>
Ämne: SV: Crazy idea or not?
Hi Christan
So, if I've understood this right, you're limited to running Windows only, and
that you are trying not to end up with the limits that Windows has as an
operating system?
It's first of all important to explain the limitations that running
Windows-only is gonna produce. Especially since the customer has a 4-hour
backup window.
As I see it, the primary idea about putting up front-end tsm servers connected
to a back-end library manager is a good idea. However, I would be running both
front-end and back-end servers on dedicated hardware with enough fibre adapters
to be able to handle the amount of I/O necessary. Usually, the issues with I/O
is not only related to Windows as operating system, but also using to less
fibre adapters / ethernet adapters or mixing random and sequential writes on
the same adapter (both has a huge impact on I/O performance). Another
consideration is the limitation on system busses. Placing all high performance
adapters on the same physical bus is not gonna improve performance either. I'm
not a Intel/Windows pro, but as far as I've seen, the scalability when it comes
to Intel machines vs for example P-machines or similiar is quite different.
A failry equipped front-end TSM server running Windows shouldnt have problems
handling +200 clients. If the issue is that you cant keep the 4-hour backup
window, it might be necessary for the customer to utilize things like LAN-free
(for non-fileservers) and subfile backups for fileservers to minimze the amount
of time required to transfer data. If the servers are fairly modern, using
compression could also bring down the amount of time required.
If none of the above helps, it's important to explain to the customer the
limitations running Windows has and that keeping that short backup-window with
that big amount of data is gonna be failry hard, or at least quite expensive
(to make up the limitations of Windows with additional hardware).
My 5 cents worth.
Daniel
________________________________________
Från: ADSM: Dist Stor Manager [ADSM-L AT VM.MARIST DOT EDU] för Christian
Svensson [Christian.Svensson AT CRISTIE DOT SE]
Skickat: den 2 februari 2009 14:45
Till: ADSM-L AT VM.MARIST DOT EDU
Ämne: SV: Crazy idea or not?
Hi Daniel,
Welcome back, longtime I heard anything from you. :D
1. Customer don't want to run any other OS then Windows.
2. Our issue today is that their Windows Servers is already have to much to do.
The I/O that TSM generates today is to much for Windows and NTFS to handle. If
we where running Linux and EXT4 then should we have much better performance and
shouldn't have this issues today.
3. Because of all the I/Os that Windows can't deal with, it automatic put the
CPU in 80-100% used. And this is why we don't have multiple instance.
4. Our goal is to get more out of the hardware and not end where Windows has it
limit.
5. End-User is today backing up more then 500+ servers to one single Windows
TSM Server, but now are we splitting up that server to 1 Library Manager and 3
front-end TSM Servers. On each server will we have max 200 servers. This is not
limited to how many TSM can handle. It is only because our backup Windows is
only Max 4 hours and each server backup in average 300 GB per night
6. This is still only a test and will not run in productions.
If I could choose I should run either Linux or AIX.
Best Regards
Christian Svensson
Cell: +46-70-325 1577
E-mail: Christian.Svensson AT cristie DOT se
Skype: cristie.christian.svensson
________________________________________
Från: ADSM: Dist Stor Manager [ADSM-L AT VM.MARIST DOT EDU] för Daniel
Sparrman [daniel.sparrman AT EXIST DOT SE]
Skickat: den 2 februari 2009 13:12
Till: ADSM-L AT VM.MARIST DOT EDU
Ämne: SV: Crazy idea or not?
Hi
First of all, i'll try to get this straight:
You want to have multiple, virtualized hosts running in a Library Manager /
Client environment.
The virtualized host will be running either on Xen or ESX? How many physical
machines are we talking about?
Since you can run multiple TSM servers on the same OS, I see no reason to
actually virtualize them. Virtualizing the TSM servers will only a) use up
resources only for the virtualization that you could have actually used for the
TSM servers instead b) Get you into trouble when it comes to sharing eventual
resources like fibre adapters, tape drives, memory, CPU.
I have one customer running dual HACMP clusters,each with 2 AIX nodes. One
HACMP cluster is running 2 TSM servers who serves around 500 clients each. The
other HACMP cluster is running another 2 TSM servers handling 500 clients each
and a 3rd TSM server only acting as the library manager. In total, we have
about 1.2PB of primary data and about 2000 clients. Each server is running on
it's own port (1500/1580, 1600/1680, 1700/1780, 1800/1880 and 1900/1980). All
these TSM servers could be running on the same node, but for security reason
(different administrators) they have been divided into 2 clusters so that they
can be managed separatly.
Could you explain more in detail what kind of environment you're actually
setting up?
Daniel
________________________________________
Från: ADSM: Dist Stor Manager [ADSM-L AT VM.MARIST DOT EDU] för Christian
Svensson [Christian.Svensson AT CRISTIE DOT SE]
Skickat: den 2 februari 2009 13:25
Till: ADSM-L AT VM.MARIST DOT EDU
Ämne: SV: Crazy idea or not?
Hi Allan,
Funny you say that I going to test BMR, but that is not true. I have moved on
from to be Mr. BMR in IBMs eyes to now work as a TSM Senoir Consultent.
The issue I have is that the customer only want to runt TSM on Windows and not
AIX or even Linux.
But to virtualize the TSM Servers we maybe can have more TSM Servers on lest
Intel Hardware. This is just a test and see what we can do.
I have recommend Linux and AIX instead but the customer only want to run
Windows Servers.
Best Regards
Christian Svensson
Cell: +46-70-325 1577
E-mail: Christian.Svensson AT cristie DOT se
Skype: cristie.christian.svensson
________________________________________
Från: ADSM: Dist Stor Manager [ADSM-L AT VM.MARIST DOT EDU] för Allen S.
Rout [asr AT UFL DOT EDU]
Skickat: den 30 januari 2009 17:44
Till: ADSM-L AT VM.MARIST DOT EDU
Ämne: Re: Crazy idea or not?
>> On Fri, 30 Jan 2009 00:00:17 +0100, Christian Svensson <Christian.Svensson
>> AT CRISTIE DOT SE> said:
> Have anyone try to setup multiple TSM Servers with Library manager
> in a VMware/XEN envirenment where are using multiple host where each
> host have, let's say 8 x 4 core CPUs with 256 GB memory in each
> host. We where thinking of to try it during Q1/Q2 this year and try
> to transfer 400 test nodes with 25-50 TB data just for fun. The
> backup will be both LAN-Free and LAN backups. But each TSM Server
> will have direct access to the tape drives via Fiber Channel.
Coming from Cristie, I'd guess you're setting up some sort of test
environment against which to run your BMR products, right? So you're
really working in a different problem space than most of us; the
tradeoffs we'd find compelling might be irrelevant to you.
I'd say running such an IO intensive app in a hardware environment
where you know you'll pay some overhead seems the wrong way to go. I
wouldn't reccomend virtualizing your TSM server unless, in your own
setting, it constitutes a relatively minor IO load.
Let's say your total load is large enough that you have multiple
64-core 256G machines to sustain it. In this case you're large enough
to deploy multiple dedicated TSM servers. You can then do your
"virtualization" within those system images. More efficient all
around, especially with memory.
But then, I'm running a 1000-node 600TB (primary) environment on a
4-core 8G P630. You're so ludicrously overpowered for the load you
describe that the virtualization overhead will be meaningless. Hell,
you could even fit Windows in there; go nuts.
- Allen S. Rout
|
