Monday, March 09, 2009

Why I chose Equallogic over NetApp

I recently had to buy some more storage for our virtualized XenServer cluster at work and just wanted to comment on how I believe NetApp have shot themselves in the foot when it comes to SMBs. I don't usually post such techie stuff, but this really annoyed me.

We currently run an older entry level NetApp FAS250 box. Its a single controller model with 14x300Gb FC disks in it and I can say it is one of the most reliable things on our network, although performance isn't incredible.

So I went to upgrade and wittled the choices down to boxes with comparible pricing: the NetApp FAS2020 (12x1TB SATA disks) and Equallogic PS5000E (16x750GB disks), both with dual controllers.

Here is a big problem with NetApp I found:

You cannot create RAID groups over 8TB on the NetApp. In addition to this, on a dual controller setup you need at least two aggregates anyway.

So, on a one box filer you not only lose a massive amount of space but also, as the more spindles you have in a RAID group the higher the performance, you lose that too!

Compare the two standard setups and see the difference:

NetApp FAS2020: 1x Five Disk RAID-DP, 1x Six Disk RAID-DP, 1x hot spare ~7TB raw space (1TB disks).

Equallogic PS5000E: 1x Fourteen disk RAID 10, 2x hot spares. ~ 5.25TB raw space (750GB disks) or 1x 15 disk RAID 50, 1x hot spare. ~7.5TB raw space.

RAID 10 is probably going to be faster than RAID DP anyway, so add the extra spindles and it should be MUCH faster than the NetApp box. Does anyone from NetApp want to comment on these stupid restrictions? I'll report more when my box arrives and tell you about the performance.

Labels: , ,


Blogger Damon Wynne said...

With comparing performance of the EQ implementation of RAID 10 to the NetApp RAID DP, I would have to say performance is pretty much a match. In some areas the EQ seems slightly ahead, such generic file copy and move tasks, but for exchange DP seems to be a bit ahead.

If the SAN isn't going to be expanded, then the RAID 10 solution is a good one, but the disk penalty after expansion seems to favour RAID DP, and if your decision is based on performance, you will be implementing storage based on smaller sized disk and more spindles as a whole. RAID 10 will really start to hurt you going down that track.

For a single shelf solution with SATA disks, I would take the cheaper FAS2020 solution personally. Who knows, the 8TB limit might be a software update and away you go...

5:21 am  
Blogger Damon Wynne said...

This comment has been removed by the author.

2:15 am  
Blogger Damon Wynne said...

Just an amendment to my earlier comment, as an option you could always go with a NetApp configuration along these lines as it turns out the 7.3.x release does give you larger raw aggregate sizes:

Upgrade to 7.3.x (16TB RAW limit for aggregates!)

RAID4 aggr0 (root) - 2 disks )one data/one parity)

RAIDDP aggr1 (rest of your storage) - 9 disks (7 data, 1 parity, 1 DP)

SPARE - 1 disk

That should net you some decent storage allocation and keeps the root aggr0 to a nice and small and hence quick rebuild.

Things to keep in mind (netapp gotchas):

1. Hotspare configuration:

options raid.min_spare_count 1

- to set just the one hot spare (Note that netapp recommends a minimum of 2 hot spares, but I think on a single shelf system you could justify a single spare)

2. Use the CLI and not the GUI to assign disks to the aggr1 - the GUI doesn't always allow to add that final last 1 or 2 disks for some reason. The CLI works everytime.

aggr add aggr1 x@ yyy

- where x is the number of disks and yyy is the size. I think 828 or something similar if you are using 1TB sata disk.

**edit - forgot to mention, obviously the RAID4 is potentially at risk from a lower level of protection than RAID DP - I think from memory, the root vol is only really around 12GB in size as a rule of thumb, so you could always copy this to the other aggr1 as backup.

2:19 am  
Blogger James White said...

"For a single shelf solution with SATA disks, I would take the cheaper FAS2020 solution personally."

If you factor in the fact that you have to pay for all the software licences on top, NetApp is far more expensive. (Snapshots, Exchange/SQL integration etc..).

Upgrade to 7.3.x (16TB RAW limit for aggregates!)

RAID4 aggr0 (root) - 2 disks )one data/one parity)

RAIDDP aggr1 (rest of your storage) - 9 disks (7 data, 1 parity, 1 DP)

SPARE - 1 disk"

Great, but with EqualLogic you don't lose 2 disks to the root partition. You can use the whole load of disks in one storage pool, again with the speed advantage of more spindles.

5:00 pm  
Blogger James White said...

Oh, lets also not forget that controllers on NetApp can't share an aggregate, so again you need two aggregates on a dual controller box anyway. Again, Equallogic wins.

5:03 pm  
Blogger Unknown said...

Hi guys,

We are using EQ 6000 for over a year as backup solution,vm,sql. Please remember what EQ is pure ISCSI NAS ,nothing else. What problems did I found:
1. Snapshots on EQ - for me it is completly disaster. If someone didn`t read documentation eq is usign 16MB block size for snapshots and 8MB for replication (as far as I remember). We did a lot of tests for exchange environemnt. Results every snaphot of datasotore=95% of his oryginal size. This mean that if your datastore is 100G and you get even one email your snapshot will be 95% or 100G. Funny:)
This happend also for vm`s running on vmware esx. Here you will loose LOT of space.

2. Free space. If you set a volume (doesn`t matter thin prov. or not) and you will fill this volume for 99% of space (simple copy files) and then you delete the data from volume (delete files from windows), eq still thinks that data are on volume. This simply means that you can free some space from host perspective but you can not free space from eq perspective, this space will never go back to free space of storage. You have to create new volume and copy data from one vol to other, You loose free space one more time.
For me EQ is great,easy to set up storage but after a year I think that speed of the array is not the way and I think we will switch to fas2020
3. Support. Well we had some "memory crc errors" ,we called support, answer "don`t worry, if you will get these errors every minute we will handle the problem".

7:16 pm  
Blogger mike said...

interesting reading and I have to agree with Damon, the EQL looks at lot better for performance than the NetApp. More spindles please.

The loss of disks for root aggr (or loss of capacity for flare) seems crazy.

Now if only Dell would give Broadcom a kick regarding drivers for their 10GbE NICs...

3:50 pm  

Post a Comment

<< Home

In Association with