--> that's because you count the fiber channel connections
In fact I count 8 SSA loops 160MB/s each. If we count FC host adapters
(max 16) we should assume >1600MB/s (16x 1063Mb/s) for F20 and twice for
800 :-)
--> but the internal busses are 4 133MB/s PCI busses
How blindish I could be sometimes. I always thought the SSA loops should
be the bottleneck so never calculated the PCI buses (the "clusters" are
H70 not H80, that's the mistake I made). But you are completely right -
buses (of H70) and SSA adapters (FC 6230) are 32-bit/33 MHz so 4 per
"cluster" will result 533 MB/s (512 MB/s is for 32 MHz ,-) or 1066 MB/s
for both when properly balanced. And because that bandwidth is for both
SSA and host attachment my numbers were two times and something over the
actual maximum. Here ought to be the source for performance improvement of
model 800!
I stand corrected. Thank you Gianluca for opening my eyes.
Zlatko Krastev
IT Consultant
Gianluca Mariani1 <gianluca_mariani AT IT.IBM DOT COM>
Sent by: "ADSM: Dist Stor Manager" <ADSM-L AT VM.MARIST DOT EDU>
24.10.2002 16:26
Please respond to "ADSM: Dist Stor Manager"
To: ADSM-L AT VM.MARIST DOT EDU
cc:
Subject: Re: RAID5 in TSM
I would just like to add something. the manuals are sometimes misleading.
ESS runs at a "theoretical" speed of 1280 MB/s yes. that's because you
count the fiber channel connections towards the outside. but the internal
busses are 4 133MB/s PCI busses. that makes a max throughput from the
cache
to the arrays of 512MB/s (I'm not talking about the new 800 which I'm not
familiar with). and those are split between the two halves of an ESS; so a
lot of care has to be taken in how you configure volumes and spread them
through the machine.
this little "trick" is what every vendor does, us (IBM), Hitachi, EMC etc.
in the same fashion you will not get the "claimed" throughput of an
Hitachi
or EMC machine in 99% of real world cases. the ESS arrays are REALLY RAID5
arrays, and the performance lag with respect to a RAID 1 machine is
negligible in the vast majority of cases. there will be those few "top
notch" customers who will not be satisfied with this, and, if they have
the
money (because RAID 5 is basically a trade off between performance and
cost), they will go for a RAID 1 machine.
I feel that saying things honestly would save us all lots of trouble and
confusion and money. this pun is obviously directed to the Sales
departments of all our companies...
Cordiali saluti
Gianluca Mariani
Tivoli TSM Global Response Team, Roma
Via Sciangai 53, Roma
phones : +39(0)659664598
+393351270554 (mobile)
gianluca_mariani AT it.ibm DOT com
----------------------------------------------------------------------------------------------------
"The people of Krikkit,are, well, you know, they're just a bunch of real
sweet guys, you know, who just happen to want to kill everybody. Hell, I
feel the same way some mornings..."
"ADSM: Dist Stor Manager" <ADSM-L AT VM.MARIST DOT EDU> wrote on 24/10/2002
14.54.40:
> Raghu,
>
> as you wrote it was already discussed on this list. In last thread we
with
> Paul confirmed both that ESS is a different beast. Very, very deep under
> the covers of ESS there are really RAID-5 arrays. But the addtional
layers
> on top of them is making ESS better than nearly all RAID-1
> implementations.
> Data taken from ESS manuals - ESS is capable to feed the data at ....
1280
> MB/s (yes, one 1,28 gigabytes per second) !!! And with new 800 model it
is
> even better.
> So answer again is "it depends". Not all RAID-5 are created equal.
> If you can afford ESS this is (IMO) the best answer for TSM.
>
> Zlatko Krastev
> IT Consultant
>
>
>
>
>
>
> Raghu S <raghu AT COSMOS.DCMDS.CO DOT IN>
> Sent by: "ADSM: Dist Stor Manager" <ADSM-L AT VM.MARIST DOT EDU>
> 23.10.2002 07:30
> Please respond to "ADSM: Dist Stor Manager"
>
>
> To: ADSM-L AT VM.MARIST DOT EDU
> cc:
> Subject: Re: RAID5 in TSM
>
>
> Paul,
>
>
> keeping TSM database,log and disk storage pool on RAID5 degrades the
> performance???
>
> Regards
>
> Raghu
>
>
>
> "Seay, Paul"
> <seay_pd@NAPT To: ADSM-L AT VM.MARIST DOT EDU
> HEON.COM> cc:
> Sent by: Subject: Re: RAID5 in TSM
> "ADSM: Dist
> Stor Manager"
> <ADSM-L AT VM DOT MA
> RIST.EDU>
>
>
> 10/22/2002
> 06:29 AM
> Please
> respond to
> "ADSM: Dist
> Stor Manager"
>
>
>
>
>
> Are you running compression?
>
> Paul D. Seay, Jr.
> Technical Specialist
> Naptheon Inc.
> 757-688-8180
>
>
> -----Original Message-----
> From: Raghu S [mailto:raghu AT COSMOS.DCMDS.CO DOT IN]
> Sent: Tuesday, October 22, 2002 8:03 AM
> To: ADSM-L AT VM.MARIST DOT EDU
> Subject: RAID5 in TSM
>
>
> Hi,
>
> There was a lot of discussion on this topic before.But i am requesting
TSM
> gurus give their comments again.
>
> The set up is like this.
>
> TSM Server : Windows NT 4.0 SP6, TSM 5.1.0.0
>
> 392 MB memory, P III
>
> Adaptech Ultra SCSI
>
> Hard Disk : Internal Hardware RAID 5:
>
> array A : 8.678GB * 3 : 17.356GB data and 8.678
> GB
> parity
>
> array B : 35.003 GB * 3 : 70.006GB data and
> 35.003
> GB parity.
>
>
> Both array A and array B are connected to the same channel.
>
> OS and TSM 5.1 are installed on array A
>
> TSM data base, recovery log and Disk storage pool are installed in array
> B.
>
> Database : 2GB+2GB = 4 GB and mirrored at TSM level on the same array
>
> Recovery Log : 500MB + 500 MB = 1 GB and mirrored at TSM level on the
same
> array
>
> Disk Storage pool : 10GB+10GB+10GB+10GB+5GB=45GB on array B
>
>
> TSM client: 4.1.2.12 ( Tivoli says 4.1.2.12 is not supported with 5.1
> Server. But i could take the backup,archive and restore with this
> combination )
>
> Number of Clients : 55, all are windows
>
> Incremental backup : 1GB/ client/day.
>
> backup window : 9AM to 6PM with 50% randamization ( all are in polling
> mode
> )
>
> LAN : 100Mbps
>
> End of the day only 10 clients could finish the backup.Remaining all are
> missing or "?" ( in progress ) or failed.
>
> Through the entire backup window the CPU load is 100% with dsmsvc.exe
> holding 98%
>
> I tested with various options. I stopped the schedular and fired 3
clients
> backup manually at the same time.Each client has 1 GB of incremental
data.
> It took three hours to finish the backup. While backing up i observed
> there
> was lot of idletime outs of sessions.
>
> Network choke is not there. I checked this with FTP.
>
> Whats the bottleneck here? Is RAID 5 is creating problems ( DB,log and
> storage pool all are on the RAID 5 )? I asked the customer to arrange a
> testing machine without any RAID. I will be getting that in two
> days.Before
> going on to the testing i like to know your comments on this.
>
>
>
> Regards
>
> Raghu S Nivas
> Consultant - TSM
> DCM Data Systems Ltd
> New Delhi
> India.
|
