New SQL Config - Hardware Proposal

Hi all,
I'm planning to install a new SQL 2005 Cluster ( Real MS Cluster ).
I wanted to know what you guys think about my config.
To roundup the charge the server will have :
- Over 6000 small databases ( with autoclose option = ON ) with a
total of 70 Gb of data
This number is so high because the main application is not
constructed well
In April of this year this number will fall to 1500 - 2000 databases
due to ( finally ) an update of this app
- +- 100 concurrent users ( via Citrix ) using this accounting
software including reporting
- Each user opens around 5 - 50 connections, sometimes I get more then
50 new connection/sec on the current system
The current system ( DL380 G3 2 * 2.8 Xeon ( 512K chache ) with 6Gb
memory ) is currently really underperforming.
2 local disks in RAID 0 as system disk and a Clariion CX3-10 ( all
fibre disks 10K ) with IPStore ( FalconStor ) in front for mirroring &
failover.
IO is no problem, Memory is fully used and CPU is avg 90 % ( with
Parallelism not active, otherwise avg 100 % and slow responses )
for the HP-minded I'm looking into following models :
- DL585 G2 with 4 procs AMD 2.8 ( dual core ) 8220SE ( too bad HP does
not have the quad-core's AMD yet )
- DL580 G5 with 4 procs Xeon 2.4Gh E7340 ( quad-core )
Is the xeon quad-core really that much faster giving a 40 % increase
in price
for the IBM-minded :
- x3850 M2 with 2 to 4 procs Xeon 2.4Gh E7330
Another question,why is there such a big price diff between HP & IBM
or should I not compare those 2 HP servers with this x3850 M2 system.
Anybody got an idea whether I should wait for the x3950 who did so
great on the TPC tests ?
Thanks in advance ...
Sven Peeters
Is your current system using AWE? You might get some CPU improvement by
going to 64bit. I would go with dual quad core 580 if I had a choice. It is
fast and leaves alot of room for growth.
Jason Massie
Web: http://statisticsio.com
RSS: http://feeds.feedburner.com/statisticsio
"Icemokka" <icemokka@.gmail.com> wrote in message
news:646e815f-a91d-47f6-9b6f-fbd01e751dc1@.e25g2000prg.googlegroups.com...
> Hi all,
> I'm planning to install a new SQL 2005 Cluster ( Real MS Cluster ).
> I wanted to know what you guys think about my config.
> To roundup the charge the server will have :
> - Over 6000 small databases ( with autoclose option = ON ) with a
> total of 70 Gb of data
> This number is so high because the main application is not
> constructed well
> In April of this year this number will fall to 1500 - 2000 databases
> due to ( finally ) an update of this app
> - +- 100 concurrent users ( via Citrix ) using this accounting
> software including reporting
> - Each user opens around 5 - 50 connections, sometimes I get more then
> 50 new connection/sec on the current system
> The current system ( DL380 G3 2 * 2.8 Xeon ( 512K chache ) with 6Gb
> memory ) is currently really underperforming.
> 2 local disks in RAID 0 as system disk and a Clariion CX3-10 ( all
> fibre disks 10K ) with IPStore ( FalconStor ) in front for mirroring &
> failover.
> IO is no problem, Memory is fully used and CPU is avg 90 % ( with
> Parallelism not active, otherwise avg 100 % and slow responses )
> for the HP-minded I'm looking into following models :
> - DL585 G2 with 4 procs AMD 2.8 ( dual core ) 8220SE ( too bad HP does
> not have the quad-core's AMD yet )
> - DL580 G5 with 4 procs Xeon 2.4Gh E7340 ( quad-core )
> Is the xeon quad-core really that much faster giving a 40 % increase
> in price
> for the IBM-minded :
> - x3850 M2 with 2 to 4 procs Xeon 2.4Gh E7330
> Another question,why is there such a big price diff between HP & IBM
> or should I not compare those 2 HP servers with this x3850 M2 system.
> Anybody got an idea whether I should wait for the x3950 who did so
> great on the TPC tests ?
>
> Thanks in advance ...
> Sven Peeters
|||DL580 with two quad-core sockets does deliver impressive compute power in my
own tests.
Linchi
"Jason Massie" wrote:

> Is your current system using AWE? You might get some CPU improvement by
> going to 64bit. I would go with dual quad core 580 if I had a choice. It is
> fast and leaves alot of room for growth.
> --
> Jason Massie
> Web: http://statisticsio.com
> RSS: http://feeds.feedburner.com/statisticsio
> "Icemokka" <icemokka@.gmail.com> wrote in message
> news:646e815f-a91d-47f6-9b6f-fbd01e751dc1@.e25g2000prg.googlegroups.com...
>
>
|||On Feb 12, 6:17Xam, Icemokka <icemo...@.gmail.com> wrote:
> Hi all,
> I'm planning to install a new SQL 2005 Cluster ( Real MS Cluster ).
> I wanted to know what you guys think about my config.
> To roundup the charge the server will have :
> - Over 6000 small databases ( with autoclose option = ON ) with a
> total of 70 Gb of data
> X This number is so high because the main application is not
> constructed well
> X In April of this year this number will fall to 1500 - 2000 databases
> due to ( finally ) an update of this app
> - +- 100 concurrent users ( via Citrix ) using this accounting
> software including reporting
> - Each user opens around 5 - 50 connections, sometimes I get more then
> 50 new connection/sec on the current system
> The current system ( DL380 G3 2 * 2.8 Xeon ( 512K chache ) with 6Gb
> memory ) is currently really underperforming.
> 2 local disks in RAID 0 as system disk and a Clariion CX3-10 ( all
> fibre disks 10K ) with IPStore (FalconStor) in front for mirroring &
> failover.
> IO is no problem, Memory is fully used and CPU is avg 90 % ( with
> Parallelism not active, otherwise avg 100 % and slow responses )
> for the HP-minded I'm looking into following models :
> - DL585 G2 with 4 procs AMD 2.8 ( dual core ) 8220SE ( too bad HP does
> not have the quad-core's AMD yet )
> - DL580 G5 with 4 procs Xeon 2.4Gh E7340 ( quad-core )
> Is the xeon quad-core really that much faster giving a 40 % increase
> in price
> for the IBM-minded :
> - x3850 M2 with 2 to 4 procs Xeon 2.4Gh E7330
> Another question,why is there such a big price diff between HP & IBM
> or should I not compare those 2 HP servers with this x3850 M2 system.
> Anybody got an idea whether I should wait for the x3950 who did so
> great on the TPC tests ?
> Thanks in advance ...
> Sven Peeters
Hi Sven.
For starters, hopefully you meant the system disks are in a Raid 1
pair not Raid 0.
I would take an Xray of your IPStor server and submit to
support@.falconstor.com for analysis. They will let you know if
something looks like it has been configured incorrectly or if there is
a patch available for any known issues. There may be some specific
settings for optimizing the Clariion that have not been done.
Your mirroring may not be done in an optimal fashion. In order to
maintain consistency between disk arrays of unmatched performance you
would have a Safe Cache area. If this is set on the same disks as your
primary storage then you're just creating twice as many writes.
Also, if you have turned on TimeMarks/SnapShots the same goes for the
SnapShot Resource Area. Due to the Copy-on-First-Write pattern you
could be really pounding away at the same disks for all three writes
plus incurring hte overhead of the extra read for each write as it
goes through the CoFW process. Ideally TimeMarks/Snapshots should be
written to an IPStor server setup in CDP mode and leave your
provisioning server to just provisioning. In the "triplet"
configuration the CDP server would hold a complete copy of your data
and then create your TimeMarks/Snapshot from there instead of your
production disks.
Lastly, double check your PCI slots and your actual IO requirements.
If you're pumping everything through a single bus you may have hit a
bottle neck. Maybe you need more ports or at least dispersed ports so
that targets and initiators are on separate HBAs.
If you're going to go ahead with new servers and are going with the
DL585 then do make sure you populate all sockets. IPStor won't need
the processing power but due to the way the AMD chipset accesses
memory you won't get full memory optimization with empty sockets. The
DL385 may be sufficient with 2 x CPUs.
Hope this helps with some new things to try.
|||On Feb 16, 12:16Xam, SnowCan...@.gmail.com wrote:
> On Feb 12, 6:17Xam, Icemokka <icemo...@.gmail.com> wrote:
>
>
>
>
>
>
> Hi Sven.
> For starters, hopefully you meant the system disks are in a Raid 1
> pair not Raid 0.
> I would take an Xray of your IPStor server and submit to
> supp...@.falconstor.com for analysis. They will let you know if
> something looks like it has been configured incorrectly or if there is
> a patch available for any known issues. There may be some specific
> settings for optimizing the Clariion that have not been done.
> Your mirroring may not be done in an optimal fashion. In order to
> maintain consistency between disk arrays of unmatched performance you
> would have a Safe Cache area. If this is set on the same disks as your
> primary storage then you're just creating twice as many writes.
> Also, if you have turned on TimeMarks/SnapShots the same goes for the
> SnapShot Resource Area. Due to the Copy-on-First-Write pattern you
> could be really pounding away at the same disks for all three writes
> plus incurring hte overhead of the extra read for each write as it
> goes through the CoFW process. Ideally TimeMarks/Snapshots should be
> written to an IPStor server setup in CDP mode and leave your
> provisioning server to just provisioning. In the "triplet"
> configuration the CDP server would hold a complete copy of your data
> and then create your TimeMarks/Snapshot from there instead of your
> production disks.
> Lastly, double check your PCI slots and your actual IO requirements.
> If you're pumping everything through a single bus you may have hit a
> bottle neck. Maybe you need more ports or at least dispersed ports so
> that targets and initiators are on separate HBAs.
> If you're going to go ahead with new servers and are going with the
> DL585 then do make sure you populate all sockets. IPStor won't need
> the processing power but due to the way the AMD chipset accesses
> memory you won't get full memory optimization with empty sockets. XThe
> DL385 may be sufficient with 2 x CPUs.
> Hope this helps with some new things to try.- Hide quoted text -
> - Show quoted text -
A further thought on this... you said you are running on a DL380 G3
which as I recall was a 32-bit platform which in turn means you are
running IPStor 4.x. Note: The final release of IPStor on 32-bit was
version 4.5 build 954 and then there have been patches and driver
updates not only to IPStor but to the OS as well. There is a
fundamental difference in how IPStor 4.x and IPStor 5.x (64-bit)
handles TimeMarks/Snapshots. In v4.x ALL TimeMarks and the resulting
CoFW blocks are held in memory until deleted. In v5.x only the LAST
TimeMark and current CoFW data is held in memory, the earlier ones are
flushed to disk. If you have EVER turned on the TimeMark feature, even
if you're not using it, then ALL writes are subject to examination and
as blacks were changed the original block is held in memory. You could
effectively be holding 100% of your disk in memory and still imposing
upto 30% performance hit watching changed blocks. If you have not
turned on TimeMarks then of course this is all irrelevant.
also, where is your mirror? You have only mentioned one IPStor server
and one Clariion. If you have virtualized the LUNs on the same
spindles you may have imposed a software Raid 1 on top of a hardware
raid 5 and then put both mirrors on the same spindles. It's only
natural that it would underperform.
BTW - that email address above should have read
support(a)falconstor.com - you may also want to check their online
knowledge base.
- Craig