The future of HAMMER1

M. L. Wilson ipc at
Sat Jul 21 02:58:51 PDT 2018


H1's "mirror-stream" is a killer feature, low overhead and just works.

In my case, compression and live deduplication were enough for me to make the switch wholesale from H1 to H2. The vast majority of my data are highly compressible and often replicated throughout thousands of directory hierarchies, so these are killer features for me. Its also nice that I don't need $3,000 worth of RAM to be able to make use of deduplication on large filesystems.

I have been using H2 even before it was declared stable, and in my experience, it is good enough to be used in production with even very demanding workloads that stress the filesystem.

While clustering is likely a ways off, some kind of mirror-stream-like functionality to cover what H1 could do I think could see many H1 users make the switch. Rsync is fine for backups as a short term fill-in, but mirror-stream is superior in many ways. The ability to have a filesystem span multiple disks is also a feature that I very much miss from H1.


On 07/21/2018 10:12 AM, Michael Neumann wrote:
> HAMMER1 was rock- solid from the early beginning (it took much less than 10
> years).  HAMMER2 is a completely different architecture, AFAICT on-media it is
> much simpler. But a lot of the complexity lies in the multi-master operation.
> I still run HAMMER1 on my server. I love "mirror-stream" for backups. That's the
> only thing I really miss from HAMMER2. But if I would setup a new server I would
> probably use HAMMER2.
> Regards,
> Michael

More information about the Users mailing list