Applcation and ZFS Access

P

pratapk

We've number of machines where native UNIX applications are running.
We've decided to try ZFS.

Can you please answer following questions,
1) Where will the ZFS server would be running, on each of these machines ?
or Do We need to allocate a different machine to run ZFS ?

2) How these each machine see the Storage, Will it see as a file server or as mounted disk ?
 
pratapk said:
1) Where will the ZFS server would be running, on each of these machines? Or do we need to allocate a different machine to run ZFS?
Click to expand...
ZFS is not a service, it's a filesystem. Just like ext3 and UFS are filesystems.

Wikipedia - ZFS


2) How these each machine see the Storage, Will it see as a file server or as mounted disk?
Click to expand...
That depends on how you export the filesystems, Samba, NFS, AFS, etc. But as with most things UNIX, it will see files.

I think you're confusing ZFS with a distributed filesystem which it is not.
 
Thanks a lot, Those are really helpful.

But, If multiple machines want to make use of storage, like a Filesystem, Not, like FileServer ( FreeNAS).
Does ZFS work ?

I want be storing user content on storage, So I've multiple applications running on multiple Servers for scaling.
I want to use mirroring/pooling to achieve good throughput from each harddrive. But, I don't much of the integrity checks.
So, If I use ZFS Storage for each machine, It would be independent to it's own.
I would not gain much by using ZFS.
 
You're probably already using NFS, this will work well with ZFS.

Create a storage server with one or more ZFS pools and export the filesystems with NFS to the other machines.
 
To have good throughput, we plan to utilize High sequential reads/writes ( at 16 MB of buffer data each).
So, We plan to use High amount of memory to buffer the data at the application level.

So, for Storage server to achieve good throughput with sequential reads/writes.
Do we need to have high amount of memory at its end too ?
How about the reliability of the storage server ?
 
If we use a centralized storage server, it may not scale to high number of concurrent users.
Is there any other storage technologies / filesystems that could help.
 
pratapk said:
To have good throughput, we plan to utilize High sequential reads/writes ( at 16 MB of buffer data each).
So, We plan to use High amount of memory to buffer the data at the application level.
Click to expand...

In that case the use of a ZIL device on SSD is highly recommended.

pratapk said:
So, for Storage server to achieve good throughput with sequential reads/writes.
Do we need to have high amount of memory at its end too ?
How about the reliability of the storage server ?
Click to expand...

You need memory on the ZFS storage server. Depending on other features you might use such as deduplication, compression, etc, you could consider the use of a separate LOG device as well.
 
Thanks gkontos. It is the most useful advice.

So far from my knowledge of ZFS, ZFS Storage is not clustered.
So, Even if we do deduplication and other features.
On 10 Gbe Network, It is limited on the overall bus bandwidth and CPU Cores.

So, the question is, Is it advisable to use single storage ?
Can it scale ?
 
pratapk said:
Thanks gkontos. It is the most useful advice.

So far from my knowledge of ZFS, ZFS Storage is not clustered.
So, Even if we do deduplication and other features.
On 10 Gbe Network, It is limited on the overall bus bandwidth and CPU Cores.

So, the question is, Is it advisable to use single storage ?
Can it scale ?
Click to expand...

For clustering you can always use HAST.

Scaling is another issue. We are currently designing a Petabyte storage solution. We are still in the design faze but I can tell you that we plan on using many JBOD Chassis driven by different servers. Our goal is to create an environment where one JBOD chassis can be controlled by more than one servers for redundancy issues.

LSI has created a nice SAS switch which we plan on utilizing.

George
 
You must log in or register to reply here.
Back
Top