No - the avoidance of SATA disks and use of only SAS <500Gb was mainly down to best practices research, but as mentioned above, cost always come in to it when choosing disks.
I did say in my first post to ignore financials, as I wanted to get a conversation based purely on disk performance, without the inclusion of $ per MB.
I recently spoke to a NetApp Partner who said he would never sell disks >500GB for virtualisation, but with the latest advancement in HDD and their onboard cache, I was wondering whether what was best practice a year ago is no longer best practice and that it need re-evaluating; i.e. 1TB are now safe to use for high work loads.
NetApp and Equallogic customer here. 1 year ago, the NetApp Partner was dealing with a technical limitation -- the amount of drives one could pool into a NetApp unit of performance ("aggregate") had to be under 16TB, at one point this was including RAID parity drives. So the most perofrmance he could sell with 1TB was at that time the equivalent of 13 1TB data disks (parity doesn't contribute to app performance). Horrible.
The issue of buying large drives is more like a problem of organizational disclipline: Can you deal with the staggering amount of space efficiently even if you run out of iops (trapped capacity)? It's that tendency to keep packing twice as much data as a 500GB drive that leads to problems, and given the above, nothing to do with the drives themselves.
"Still holds true... " - is this aimed whether it is safe to use 1TB disks or whether using SATA for virtualisation
My test results show that SATAs would suffice, my question is more aimed at whether 1TB disks are safe to use or whether the seek times would be too large...
Still holds true, that best practices states that seek times can be to long on large slow drives for many applications, not just virtualization, but virtualization could certainly be included as one of those applications.
Now the reality? I see large slow sata disk arrays running very large data centers where hosting companies are charging to host SQL, Exchange, and every other application under the sun. Right now? I am running my heaviest I/O VM's on a tray of big slow 1TB sata drives and it's running fine. I am running the rest of the enterprise on a tray of 1TB 7.2K Nearline SAS, but that is what they alwasy ran on. :) (long story on why we are runnign this way right now)
Still holds true... But I can assure you thousands of sans are out there right now running all sorts of loads on crud SATA drives. You should be fine, but the first signs of bad performance and support is going to point squarely at your 7.2K drives as the issue. I have been there done that. Once you really figure out what is going on, you realize it had nothing to do with the drives in the first place.
Now the reality? I see large slow sata disk arrays running very large data centers where hosting companies are charging to host SQL, Exchange, and every other application under the sun. Right now? I am running my heaviest I/O VM's on a tray of big slow 1TB sata drives and it's running fine. I am running the rest of the enterprise on a tray of 1TB 7.2K Nearline SAS, but that is what they alwasy ran on. :) (long story on why we are runnign this way right now)
We run a significant chunk of our 'enterprise' on 3x PS6500E with 1TB sata drives. This includes more than 40 virtual servers, sharepoint databases, soon to be Exchange 2010.
It's all about what we can afford and what our users will find acceptable.
Few shops are lucky to be able to afford their peak workloads at minimal latencies 100% of the time. As long as our avg latencies aren't too high, our users are fine....
Yep..... As I said, the real bummer is you call support and have some performance issues and they point right at the slow, large drives is the real problem even though the issue probably has nothing to do with the drives. We have some nice fast SAS drives we will be moving much of the load to soon. We are putting in Storage Virtualization servers, and are sort of stuck in between righ tnow. But we run a lot of stuff on 1TG Nearline SAS by design.
Yep..... As I said, the real bummer is you call support and have some performance issues and they point right at the slow, large drives is the real problem even though the issue probably has nothing to do with the drives.
you should be able to point to your low SANHQ latencies and the slow drives won't be a support issue.
Latency's aren't all that low though.... We stress 15 1TB SATA spindles drives pretty good. We really need to move stuff around. Soon..... Or so I have been promised it will be soon anyways.
Never say never.... :) It really does come down to needed IOPS here.... If the spindles can deliver the needed IOPS nobody would know what drives it's running on. There are 600GB 15K sas drives that make his statement silly in all honesty as those drives are as fast as it gets period.
No - the avoidance of SATA disks and use of only SAS <500Gb was mainly down to best practices research, but as mentioned above, cost always come in to it when choosing disks.
I did say in my first post to ignore financials, as I wanted to get a conversation based purely on disk performance, without the inclusion of $ per MB.
I recently spoke to a NetApp Partner who said he would never sell disks >500GB for virtualisation, but with the latest advancement in HDD and their onboard cache, I was wondering whether what was best practice a year ago is no longer best practice and that it need re-evaluating; i.e. 1TB are now safe to use for high work loads.
I did some research last night, and looked up seek times/latency times on some of the major HDD vendors last night and the times were similair for both 500GB and 1TB - although I don't know if this is just laziness on their part to put the correct data on for each disk size.
If anything, we will be getting nearline SAS, and not plain SATA. I am also asking vendors for quotes on 24x1TB vs 48x500GB and seeing what differences there are in price. The 500GB will be a lot cheaper, but more disks = more chassises to house them.
evilensky
1 Rookie
•
74 Posts
1
October 4th, 2011 17:00
NetApp and Equallogic customer here. 1 year ago, the NetApp Partner was dealing with a technical limitation -- the amount of drives one could pool into a NetApp unit of performance ("aggregate") had to be under 16TB, at one point this was including RAID parity drives. So the most perofrmance he could sell with 1TB was at that time the equivalent of 13 1TB data disks (parity doesn't contribute to app performance). Horrible.
The issue of buying large drives is more like a problem of organizational disclipline: Can you deal with the staggering amount of space efficiently even if you run out of iops (trapped capacity)? It's that tendency to keep packing twice as much data as a 500GB drive that leads to problems, and given the above, nothing to do with the drives themselves.
RichardM82
4 Posts
0
October 4th, 2011 13:00
Hi John,
THanks for the response.
"Still holds true... " - is this aimed whether it is safe to use 1TB disks or whether using SATA for virtualisation
My test results show that SATAs would suffice, my question is more aimed at whether 1TB disks are safe to use or whether the seek times would be too large...
JOHNADCO
2 Intern
•
847 Posts
0
October 4th, 2011 13:00
Still holds true, that best practices states that seek times can be to long on large slow drives for many applications, not just virtualization, but virtualization could certainly be included as one of those applications.
Now the reality? I see large slow sata disk arrays running very large data centers where hosting companies are charging to host SQL, Exchange, and every other application under the sun. Right now? I am running my heaviest I/O VM's on a tray of big slow 1TB sata drives and it's running fine. I am running the rest of the enterprise on a tray of 1TB 7.2K Nearline SAS, but that is what they alwasy ran on. :) (long story on why we are runnign this way right now)
JOHNADCO
2 Intern
•
847 Posts
0
October 4th, 2011 13:00
Still holds true... But I can assure you thousands of sans are out there right now running all sorts of loads on crud SATA drives. You should be fine, but the first signs of bad performance and support is going to point squarely at your 7.2K drives as the issue. I have been there done that. Once you really figure out what is going on, you realize it had nothing to do with the drives in the first place.
evilensky
1 Rookie
•
74 Posts
0
October 4th, 2011 14:00
We run a significant chunk of our 'enterprise' on 3x PS6500E with 1TB sata drives. This includes more than 40 virtual servers, sharepoint databases, soon to be Exchange 2010.
It's all about what we can afford and what our users will find acceptable.
Few shops are lucky to be able to afford their peak workloads at minimal latencies 100% of the time. As long as our avg latencies aren't too high, our users are fine....
evilensky
1 Rookie
•
74 Posts
0
October 4th, 2011 14:00
Do you have a link to any manufacturer specifications where the seek time for a 1TB is larger than for an equivalent model 500GB drive?
JOHNADCO
2 Intern
•
847 Posts
0
October 4th, 2011 14:00
Yep..... As I said, the real bummer is you call support and have some performance issues and they point right at the slow, large drives is the real problem even though the issue probably has nothing to do with the drives. We have some nice fast SAS drives we will be moving much of the load to soon. We are putting in Storage Virtualization servers, and are sort of stuck in between righ tnow. But we run a lot of stuff on 1TG Nearline SAS by design.
evilensky
1 Rookie
•
74 Posts
0
October 4th, 2011 14:00
you should be able to point to your low SANHQ latencies and the slow drives won't be a support issue.
JOHNADCO
2 Intern
•
847 Posts
0
October 4th, 2011 15:00
Latency's aren't all that low though.... We stress 15 1TB SATA spindles drives pretty good. We really need to move stuff around. Soon..... Or so I have been promised it will be soon anyways.
JOHNADCO
2 Intern
•
847 Posts
0
October 4th, 2011 16:00
Never say never.... :) It really does come down to needed IOPS here.... If the spindles can deliver the needed IOPS nobody would know what drives it's running on. There are 600GB 15K sas drives that make his statement silly in all honesty as those drives are as fast as it gets period.
RichardM82
4 Posts
0
October 4th, 2011 16:00
No - the avoidance of SATA disks and use of only SAS <500Gb was mainly down to best practices research, but as mentioned above, cost always come in to it when choosing disks.
I did say in my first post to ignore financials, as I wanted to get a conversation based purely on disk performance, without the inclusion of $ per MB.
I recently spoke to a NetApp Partner who said he would never sell disks >500GB for virtualisation, but with the latest advancement in HDD and their onboard cache, I was wondering whether what was best practice a year ago is no longer best practice and that it need re-evaluating; i.e. 1TB are now safe to use for high work loads.
RichardM82
4 Posts
0
October 5th, 2011 12:00
Thanks everyone for your input - very useful.
I did some research last night, and looked up seek times/latency times on some of the major HDD vendors last night and the times were similair for both 500GB and 1TB - although I don't know if this is just laziness on their part to put the correct data on for each disk size.
If anything, we will be getting nearline SAS, and not plain SATA. I am also asking vendors for quotes on 24x1TB vs 48x500GB and seeing what differences there are in price. The 500GB will be a lot cheaper, but more disks = more chassises to house them.