Hi all Is there anything usable for zfs/openindiana that allows for multi-tiered storage? The scenario is as thus: We have a 50TB storage unit which was built to be an archive, but lately, scientists have been using this for a fileserver for modelling. Pracitaclly, this means 50+ processes doing more or less random i/o to the server, which a 4-VDEV RAIDz2 system isn''t very well suited to handle. Since we don''t have sufficient storage to move the data they are modelling, we consider getting a bunch of SSDs for L2ARC/SLOG, which may solve the problem. But then, a better solution might be a two-tier storage solution, with fast drives on a server in front and slowish drives in the back. Such systems exists, but most of them are commercially built, meaning the price per terabyte will be about 5-10x the price of using ZFS on an open system. Anyone here that know such a system? Vennlige hilsener / Best regards roy -- Roy Sigurd Karlsbakk (+47) 97542685 roy at karlsbakk.net http://blogg.karlsbakk.net/ -- I all pedagogikk er det essensielt at pensum presenteres intelligibelt. Det er et element?rt imperativ for alle pedagoger ? unng? eksessiv anvendelse av idiomer med fremmed opprinnelse. I de fleste tilfeller eksisterer adekvate og relevante synonymer p? norsk.
> > Hi all > > > > Is there anything usable for zfs/openindiana that allows for > > multi-tiered storage? > > > > The scenario is as thus: We have a 50TB storage unit which was built > > to be an archive, but lately, scientists have been using this for a > > fileserver for modelling. Pracitaclly, this means 50+ processes > > doing more or less random i/o to the server, which a 4-VDEV RAIDz2 > > system isn''t very well suited to handle. Since we don''t have > > sufficient storage to move the data they are modelling, we consider > > getting a bunch of SSDs for L2ARC/SLOG, which may solve the problem. > > But then, a better solution might be a two-tier storage solution, > > with fast drives on a server in front and slowish drives in the > > back. Such systems exists, but most of them are commercially built, > > meaning the price per terabyte will be about 5-10x the price of > > using ZFS on an open system. > > This sounds almost exactly like the sort of problem SSDs and > SLOG/L2ARC are designed for.Not really, no. SLOG/L2ARC is fine for small amounts of data, but if you have 50TB or even more, where perhaps 10% are used more or less regularly, then those SSDs are likely to cost a wee bit. Tiered storage or hierarchical storage management ensures that data is available, but not necessarily from the fastest medium. I remember Novell had a solution for this using tape, and although that''s not really current, it''s the same idea. What I''m looking for is something like this: Slow server in the back with 50-100TB or so, RAIDz2 on lots of 7k2 drives. Fast server in the front with, say, 5-10TB on striped mirrores of 15k drives. But then again, I just wrote a spreadsheet to check how much it''ll cost for 5TB on a dedicated server on 15k drives and compared that to 5TB worth of L2ARC on Crucial^WMicron RealSSD 256MB, and found the two to be so close in cost, they''re basically equal, so I guess I''ll rest my case on this :P Vennlige hilsener / Best regards roy -- Roy Sigurd Karlsbakk (+47) 97542685 roy at karlsbakk.net http://blogg.karlsbakk.net/ -- I all pedagogikk er det essensielt at pensum presenteres intelligibelt. Det er et element?rt imperativ for alle pedagoger ? unng? eksessiv anvendelse av idiomer med fremmed opprinnelse. I de fleste tilfeller eksisterer adekvate og relevante synonymer p? norsk.
On Jan 26, 2011, at 19:48, Roy Sigurd Karlsbakk wrote:> The scenario is as thus: We have a 50TB storage unit which was built to be an archive, but lately, scientists have been using this for a fileserver for modelling. Pracitaclly, this means 50+ processes doing more or less random i/o to the server, which a 4-VDEV RAIDz2 system isn''t very well suited to handle. [...] > > Anyone here that know such a system?Isn''t Lustre designed to handle these situations? You can have n-number of back-end file servers where I/O is distributed, but it all shows up in one namespace / mount-point. You can then have multiple mount-points with different I/O characteristics: /scratch is smaller, but on striped-mirrors (one set of server/s); /home or /data is larger, but on slower more space efficient RAID (possible another set of server/s, or a different FS on the same set as above). http://en.wikipedia.org/wiki/Lustre_(file_system) Sun/Oracle has their SAM-QFS combination if you want hierarchical storage management: http://en.wikipedia.org/wiki/QFS This isn''t wholly a technical problem either: the scientists need to have a place to put ''scratch'' work for a few days?and be encouraged to use it?and only then move the data to the archive area. They don''t seem to have the resources available (welcome to academia), and so are making due with what is actually there.