Which is fastest, and which is fast enough?
We are struggling with one 3rd party supplier that insist that our iSCSI solution is around 5000% slower than our previous solution. They have also stated that they would not support any speed related issues with the product if we moved the DB on to iSCSI.
No just to fill a few details, the old storage was a HP EVA 8000, a big beastie if ever there was one and the current is a Dell EqualLogic which is more utilitarian to say the least.
Now anyone with half a brain cell will say the EVA is going to be quicker than the EQ. Any test you throw at the the two solutions will back that up to varying degrees of excess.
Now everyone accepts that, so where may you ask is the problem?
Well, what we should be asking is not how fast is our choice of storage but is it fast enough and how does it affect the end user experience?
If the storage is runs at half the speed but this represents only 1% of the time an application takes to respond to the user then to be honest the user is not really going to notice much change.
This is where it is important to be able to test and replicate the user experience rather than just throw a meaningless artificial test like IOMeter at the problem. This requires a whole heap of application profiling and a testing methodology that does not come cheap but if you can get the buy in of the developer it can work wonders on improving performance and tuning applications for the real world of the dreaded user.
Our EVA is not replicated to our DR site as the license costs are prohibitive, it is six years old now and costs us £12k a year in maintenance. The only app on it is this one, it is important but at the end of the day our sap boxes are visualized and run on iSCSI, our Data warehouse runs on iSCSI and all our other 150 servers run on iSCSI including SQL and Exchange and 95% are visualized as well.
Unfortunately this is a software company that specified all the hardware for one project before they had even written a line of code let alone done any kind of performance testing. The main server was specced at 4 sockets and 24 cores with 64GB of RAM. At the time this was twice as big as any of our VMware hosts that run all our other servers.
Developers don't you just love them.
No just to fill a few details, the old storage was a HP EVA 8000, a big beastie if ever there was one and the current is a Dell EqualLogic which is more utilitarian to say the least.
Now anyone with half a brain cell will say the EVA is going to be quicker than the EQ. Any test you throw at the the two solutions will back that up to varying degrees of excess.
Now everyone accepts that, so where may you ask is the problem?
Well, what we should be asking is not how fast is our choice of storage but is it fast enough and how does it affect the end user experience?
If the storage is runs at half the speed but this represents only 1% of the time an application takes to respond to the user then to be honest the user is not really going to notice much change.
This is where it is important to be able to test and replicate the user experience rather than just throw a meaningless artificial test like IOMeter at the problem. This requires a whole heap of application profiling and a testing methodology that does not come cheap but if you can get the buy in of the developer it can work wonders on improving performance and tuning applications for the real world of the dreaded user.
Our EVA is not replicated to our DR site as the license costs are prohibitive, it is six years old now and costs us £12k a year in maintenance. The only app on it is this one, it is important but at the end of the day our sap boxes are visualized and run on iSCSI, our Data warehouse runs on iSCSI and all our other 150 servers run on iSCSI including SQL and Exchange and 95% are visualized as well.
Unfortunately this is a software company that specified all the hardware for one project before they had even written a line of code let alone done any kind of performance testing. The main server was specced at 4 sockets and 24 cores with 64GB of RAM. At the time this was twice as big as any of our VMware hosts that run all our other servers.
Developers don't you just love them.
Comments
Post a Comment