Hi all;
Am trying to understand what sort of performance difference I might see between two different configurations:
1) IBM N6240 E21 (FAS3240C) w/ 120x600GB 15K 3.5" SAS and 512GB of flash cache 2) IBM N6250 E26 w/ 80x900GB 10K 2.5" SAS and 512GB of flash cache.
Sorry, on the latter I don't know the equivalent FAS. Probably FAS3250C?
We have fewer spindles, but newer, beefier controllers.
Our workload is primarily VMware via NFS. Lotsa random reads and writes (more on the read side) with I'd say the bulk of the IO requests in the 64KB+ range.
Will I regret going with fewer spindles?
Ray
I'm usually in the camp of more spindles is better then not enough. Spindles tend to dictate performance more then the controller does, in most cases, but not all.
Typed with my thumbs!
On Sep 13, 2013, at 12:28 PM, Ray Van Dolson rvandolson@esri.com wrote:
Hi all;
Am trying to understand what sort of performance difference I might see between two different configurations:
- IBM N6240 E21 (FAS3240C) w/ 120x600GB 15K 3.5" SAS and 512GB of
flash cache 2) IBM N6250 E26 w/ 80x900GB 10K 2.5" SAS and 512GB of flash cache.
Sorry, on the latter I don't know the equivalent FAS. Probably FAS3250C?
We have fewer spindles, but newer, beefier controllers.
Our workload is primarily VMware via NFS. Lotsa random reads and writes (more on the read side) with I'd say the bulk of the IO requests in the 64KB+ range.
Will I regret going with fewer spindles?
Ray _______________________________________________ Toasters mailing list Toasters@teaparty.net http://www.teaparty.net/mailman/listinfo/toasters
The capacity is about a wash. You will get 2x the IOP with your configuration, but you will save in U space with the 10ks. Personally, I agree with "thelastman"... Oh, and flash is your friend!
From: thelastman@gmail.com Subject: Re: Expected performance difference between two configurations Date: Fri, 13 Sep 2013 12:36:24 -0700 To: rvandolson@esri.com CC: toasters@teaparty.net
I'm usually in the camp of more spindles is better then not enough. Spindles tend to dictate performance more then the controller does, in most cases, but not all.
Typed with my thumbs!
On Sep 13, 2013, at 12:28 PM, Ray Van Dolson rvandolson@esri.com wrote:
Hi all;
Am trying to understand what sort of performance difference I might see between two different configurations:
- IBM N6240 E21 (FAS3240C) w/ 120x600GB 15K 3.5" SAS and 512GB of
flash cache 2) IBM N6250 E26 w/ 80x900GB 10K 2.5" SAS and 512GB of flash cache.
Sorry, on the latter I don't know the equivalent FAS. Probably FAS3250C?
We have fewer spindles, but newer, beefier controllers.
Our workload is primarily VMware via NFS. Lotsa random reads and writes (more on the read side) with I'd say the bulk of the IO requests in the 64KB+ range.
Will I regret going with fewer spindles?
Ray _______________________________________________ Toasters mailing list Toasters@teaparty.net http://www.teaparty.net/mailman/listinfo/toasters
Toasters mailing list Toasters@teaparty.net http://www.teaparty.net/mailman/listinfo/toasters
How dynamic is your environment? It is very simple to add shelves in the future as your load requirements increase. It is a bit more complex, and costly, to do a controller upgrade. A few things to consider.
Controller: The FAS3250 represents the current generation of NetApp controllers, and if you aren't considering Clustered Data OnTap, you should be. The 3240 can only participate in a cluster with 4 or fewer nodes. The 3250 can go to 8 nodes (62xx series can do 24). There are other improvements you get in a 3250, but essentially you are buying a previous generation to save $ with the 3240.
Shelves and Drives: NetApp disk shelves (DS2246, DS4243, and DS4246) all have 24 drives. Your config uses multiples of 20 making me question how. You are also comparing 600gb to 900gb drives SAS drives, not just 10k 2.5" vs 15k 3.5" drives. That's 2 factors slowing you down.
You can buy 600, 900, and now 1200 gb SAS 2.5" 10k RPM disks. These all have similar performance profiles, but as spindle size increases your iops / GB drop. You also should compare prices on these shelves and see that the larger sizes cost more per GB than the smaller ones.
Note that NetApp has EOA'd (End of Availability) the DS4243 shelves with SAS 3.5" 15k drives. Buying these now means that your next disk purchase will end up being different hardware. You do get more IOPS from the 15k drives, but you pay in power, space, and buying hardware that's no longer being sold by NetApp.
I suspect that if you have this re-quoted to have the 3240 to 3250 comparison with the same drive and shelf type (I'd go with the DS2246 600gb drives but you can look at the 900s on both too) you'll see less spread in price and your choice will be easier.
Finally a question for you. Why would you buy IBM rebranded filers and not NetApp? I am curious as to what the IBM value add is in the equation?
Colin Bieberstein
On 2013-09-13, at 1:36 PM, Blake Golliher <@gmail.com> wrote:
I'm usually in the camp of more spindles is better then not enough. Spindles tend to dictate performance more then the controller does, in most cases, but not all.
Typed with my thumbs!
On Sep 13, 2013, at 12:28 PM, Ray Van Dolson rvandolson@esri.com wrote:
Hi all;
Am trying to understand what sort of performance difference I might see between two different configurations:
- IBM N6240 E21 (FAS3240C) w/ 120x600GB 15K 3.5" SAS and 512GB of
flash cache 2) IBM N6250 E26 w/ 80x900GB 10K 2.5" SAS and 512GB of flash cache.
Sorry, on the latter I don't know the equivalent FAS. Probably FAS3250C?
We have fewer spindles, but newer, beefier controllers.
Our workload is primarily VMware via NFS. Lotsa random reads and writes (more on the read side) with I'd say the bulk of the IO requests in the 64KB+ range.
Will I regret going with fewer spindles?
Ray _______________________________________________ Toasters mailing list Toasters@teaparty.net http://www.teaparty.net/mailman/listinfo/toasters
Toasters mailing list Toasters@teaparty.net http://www.teaparty.net/mailman/listinfo/toasters
Is NetApp no longer selling 15k RPM drives at all? They were really pushing 15K SAS as "the answer" not that long ago - so it seems weird that now everything is being quoted with 10K 2.5" SAS instead.
Davin.
On Fri, Sep 13, 2013 at 9:40 PM, Colin Bieberstein colin@bieberstein.cawrote:
How dynamic is your environment? It is very simple to add shelves in the future as your load requirements increase. It is a bit more complex, and costly, to do a controller upgrade. A few things to consider.
Controller: The FAS3250 represents the current generation of NetApp controllers, and if you aren't considering Clustered Data OnTap, you should be. The 3240 can only participate in a cluster with 4 or fewer nodes. The 3250 can go to 8 nodes (62xx series can do 24). There are other improvements you get in a 3250, but essentially you are buying a previous generation to save $ with the 3240.
Shelves and Drives: NetApp disk shelves (DS2246, DS4243, and DS4246) all have 24 drives. Your config uses multiples of 20 making me question how. You are also comparing 600gb to 900gb drives SAS drives, not just 10k 2.5" vs 15k 3.5" drives. That's 2 factors slowing you down.
You can buy 600, 900, and now 1200 gb SAS 2.5" 10k RPM disks. These all have similar performance profiles, but as spindle size increases your iops / GB drop. You also should compare prices on these shelves and see that the larger sizes cost more per GB than the smaller ones.
Note that NetApp has EOA'd (End of Availability) the DS4243 shelves with SAS 3.5" 15k drives. Buying these now means that your next disk purchase will end up being different hardware. You do get more IOPS from the 15k drives, but you pay in power, space, and buying hardware that's no longer being sold by NetApp.
I suspect that if you have this re-quoted to have the 3240 to 3250 comparison with the same drive and shelf type (I'd go with the DS2246 600gb drives but you can look at the 900s on both too) you'll see less spread in price and your choice will be easier.
Finally a question for you. Why would you buy IBM rebranded filers and not NetApp? I am curious as to what the IBM value add is in the equation?
Colin Bieberstein
On 2013-09-13, at 1:36 PM, Blake Golliher <@gmail.com> wrote:
I'm usually in the camp of more spindles is better then not enough.
Spindles tend to dictate performance more then the controller does, in most cases, but not all.
Typed with my thumbs!
On Sep 13, 2013, at 12:28 PM, Ray Van Dolson rvandolson@esri.com
wrote:
Hi all;
Am trying to understand what sort of performance difference I might see between two different configurations:
- IBM N6240 E21 (FAS3240C) w/ 120x600GB 15K 3.5" SAS and 512GB of
flash cache 2) IBM N6250 E26 w/ 80x900GB 10K 2.5" SAS and 512GB of flash cache.
Sorry, on the latter I don't know the equivalent FAS. Probably FAS3250C?
We have fewer spindles, but newer, beefier controllers.
Our workload is primarily VMware via NFS. Lotsa random reads and writes (more on the read side) with I'd say the bulk of the IO requests in the 64KB+ range.
Will I regret going with fewer spindles?
Ray _______________________________________________ Toasters mailing list Toasters@teaparty.net http://www.teaparty.net/mailman/listinfo/toasters
Toasters mailing list Toasters@teaparty.net http://www.teaparty.net/mailman/listinfo/toasters
Toasters mailing list Toasters@teaparty.net http://www.teaparty.net/mailman/listinfo/toasters
On Tue, Sep 17, 2013 at 12:55:26PM -0400, Davin Milun wrote:
Is NetApp no longer selling 15k RPM drives at all? They were really pushing 15K SAS as "the answer" not that long ago - so it seems weird that now everything is being quoted with 10K 2.5" SAS instead.
Davin.
That's my understanding. We were told it was an "industry" thing -- precipitated by SSD/flash.
Ray
On Fri, Sep 13, 2013 at 9:40 PM, Colin Bieberstein colin@bieberstein.ca wrote:
How dynamic is your environment? It is very simple to add shelves in the future as your load requirements increase. It is a bit more complex, and costly, to do a controller upgrade. A few things to consider. Controller: The FAS3250 represents the current generation of NetApp controllers, and if you aren't considering Clustered Data OnTap, you should be. The 3240 can only participate in a cluster with 4 or fewer nodes. The 3250 can go to 8 nodes (62xx series can do 24). There are other improvements you get in a 3250, but essentially you are buying a previous generation to save $ with the 3240. Shelves and Drives: NetApp disk shelves (DS2246, DS4243, and DS4246) all have 24 drives. Your config uses multiples of 20 making me question how. You are also comparing 600gb to 900gb drives SAS drives, not just 10k 2.5" vs 15k 3.5" drives. That's 2 factors slowing you down. You can buy 600, 900, and now 1200 gb SAS 2.5" 10k RPM disks. These all have similar performance profiles, but as spindle size increases your iops / GB drop. You also should compare prices on these shelves and see that the larger sizes cost more per GB than the smaller ones. Note that NetApp has EOA'd (End of Availability) the DS4243 shelves with SAS 3.5" 15k drives. Buying these now means that your next disk purchase will end up being different hardware. You do get more IOPS from the 15k drives, but you pay in power, space, and buying hardware that's no longer being sold by NetApp. I suspect that if you have this re-quoted to have the 3240 to 3250 comparison with the same drive and shelf type (I'd go with the DS2246 600gb drives but you can look at the 900s on both too) you'll see less spread in price and your choice will be easier. Finally a question for you. Why would you buy IBM rebranded filers and not NetApp? I am curious as to what the IBM value add is in the equation? Colin Bieberstein On 2013-09-13, at 1:36 PM, Blake Golliher <@gmail.com> wrote: > I'm usually in the camp of more spindles is better then not enough. Spindles tend to dictate performance more then the controller does, in most cases, but not all. > > Typed with my thumbs! > >> On Sep 13, 2013, at 12:28 PM, Ray Van Dolson <rvandolson@esri.com> wrote: >> >> Hi all; >> >> Am trying to understand what sort of performance difference I might see >> between two different configurations: >> >> 1) IBM N6240 E21 (FAS3240C) w/ 120x600GB 15K 3.5" SAS and 512GB of >> flash cache >> 2) IBM N6250 E26 w/ 80x900GB 10K 2.5" SAS and 512GB of flash cache. >> >> Sorry, on the latter I don't know the equivalent FAS. Probably >> FAS3250C? >> >> We have fewer spindles, but newer, beefier controllers. >> >> Our workload is primarily VMware via NFS. Lotsa random reads and >> writes (more on the read side) with I'd say the bulk of the IO requests >> in the 64KB+ range. >> >> Will I regret going with fewer spindles? >> >> Ray
Over simplified
120*200 = 24,000 maximum sustained ops (minus parity disks)
Or
80*140 = 11,200 maximum sustained ops (minus parity disks)
-----Original Message----- From: toasters-bounces@teaparty.net [mailto:toasters-bounces@teaparty.net] On Behalf Of Ray Van Dolson Sent: Friday, September 13, 2013 3:29 PM To: toasters@teaparty.net Subject: Expected performance difference between two configurations
Hi all;
Am trying to understand what sort of performance difference I might see between two different configurations:
1) IBM N6240 E21 (FAS3240C) w/ 120x600GB 15K 3.5" SAS and 512GB of flash cache 2) IBM N6250 E26 w/ 80x900GB 10K 2.5" SAS and 512GB of flash cache.
Sorry, on the latter I don't know the equivalent FAS. Probably FAS3250C?
We have fewer spindles, but newer, beefier controllers.
Our workload is primarily VMware via NFS. Lotsa random reads and writes (more on the read side) with I'd say the bulk of the IO requests in the 64KB+ range.
Will I regret going with fewer spindles?
Ray _______________________________________________ Toasters mailing list Toasters@teaparty.net http://www.teaparty.net/mailman/listinfo/toasters
[insert the 'what is an op' argument]
Apps don't talk disk ops NFS doesn't equate to disk ops ... ...
Sent from my iPhone
On Sep 13, 2013, at 2:37 PM, Jordan Slingerland Jordan.Slingerland@independenthealth.com wrote:
Over simplified
120*200 = 24,000 maximum sustained ops (minus parity disks)
Or
80*140 = 11,200 maximum sustained ops (minus parity disks)
-----Original Message----- From: toasters-bounces@teaparty.net [mailto:toasters-bounces@teaparty.net] On Behalf Of Ray Van Dolson Sent: Friday, September 13, 2013 3:29 PM To: toasters@teaparty.net Subject: Expected performance difference between two configurations
Hi all;
Am trying to understand what sort of performance difference I might see between two different configurations:
- IBM N6240 E21 (FAS3240C) w/ 120x600GB 15K 3.5" SAS and 512GB of
flash cache 2) IBM N6250 E26 w/ 80x900GB 10K 2.5" SAS and 512GB of flash cache.
Sorry, on the latter I don't know the equivalent FAS. Probably FAS3250C?
We have fewer spindles, but newer, beefier controllers.
Our workload is primarily VMware via NFS. Lotsa random reads and writes (more on the read side) with I'd say the bulk of the IO requests in the 64KB+ range.
Will I regret going with fewer spindles?
Ray _______________________________________________ Toasters mailing list Toasters@teaparty.net http://www.teaparty.net/mailman/listinfo/toasters
Toasters mailing list Toasters@teaparty.net http://www.teaparty.net/mailman/listinfo/toasters
I've also seen some anecdotal evidence to suggest that the real world performance between 15k and 10k is actually pretty close to even, so I'm not so sure you're going to get all that much of a difference out of the speed alone. The number of spindles, though, will certainly make a some difference. In general, the wider you can stripe, the better.
-Adam
On Fri, Sep 13, 2013 at 4:26 PM, Jeff Mohler speedtoys.racing@gmail.comwrote:
[insert the 'what is an op' argument]
Apps don't talk disk ops NFS doesn't equate to disk ops ... ...
Sent from my iPhone
On Sep 13, 2013, at 2:37 PM, Jordan Slingerland < Jordan.Slingerland@independenthealth.com> wrote:
Over simplified
120*200 = 24,000 maximum sustained ops (minus parity disks)
Or
80*140 = 11,200 maximum sustained ops (minus parity disks)
-----Original Message----- From: toasters-bounces@teaparty.net [mailto:
toasters-bounces@teaparty.net] On Behalf Of Ray Van Dolson
Sent: Friday, September 13, 2013 3:29 PM To: toasters@teaparty.net Subject: Expected performance difference between two configurations
Hi all;
Am trying to understand what sort of performance difference I might see
between two different configurations:
- IBM N6240 E21 (FAS3240C) w/ 120x600GB 15K 3.5" SAS and 512GB of
flash cache 2) IBM N6250 E26 w/ 80x900GB 10K 2.5" SAS and 512GB of flash cache.
Sorry, on the latter I don't know the equivalent FAS. Probably FAS3250C?
We have fewer spindles, but newer, beefier controllers.
Our workload is primarily VMware via NFS. Lotsa random reads and writes
(more on the read side) with I'd say the bulk of the IO requests in the 64KB+ range.
Will I regret going with fewer spindles?
Ray _______________________________________________ Toasters mailing list Toasters@teaparty.net http://www.teaparty.net/mailman/listinfo/toasters
Toasters mailing list Toasters@teaparty.net http://www.teaparty.net/mailman/listinfo/toasters
Toasters mailing list Toasters@teaparty.net http://www.teaparty.net/mailman/listinfo/toasters