Windows Server 2012

[robnicholson]

We're currently designing a new infrastructure with new StarWind SANs running on Windows Server 2012. Some questions:

• Windows Server 2012 improves network teaming. Our SAN will have six NICs with five being dedicated to the server backbone and the sixth for the general LAN. We have the option to use either iSCSI multipath or Windows Server 2012 network teaming. Both will bring NIC redundancy and spread the load. Any comments on which is preferred?
• Windows Server 2012 introduces resilient file system. The name make it sound very attractive for a SAN environment where resilience is critical. However, it's a) a new technology and b) it has some limitations (like no sparse files). Any thoughts on whether or not it's a bad idea for the host SAN drives?
• Any other Windows Server 2012 comments/suggestions/recommendations/best practises around StarWind?

Cheers, Rob.

[anton (staff)]

1) NEVER use NIC teaming with iSCSI. It's a very questionable feature for SMB (runs good on paper and with a huge load tests "see how fat pipe we have now")
but screws up everything with pulsating and small packet traffic because of an enormous latency increase. With iSCSI MPIO (RR) is a way to go. It's not
StarWind it's general requirement for block level transfers (even for FCoE and AoE I hate so much).

2) ReFS is a disaster. It's basically "good" old NTFS with some old features removed (streams, symlinks), couple of new being incompatible as concurrent teams
did no manage to have a deal (no dedupe which sucks badly) and untouched performance and design (no block level cache, CSV cache works at much higher
level and for CSV volumes only, no log structuring even as an acceleration technique etc). So the only usable feature of ReFS is digital signatures so consumer
grade SATA drives will not spoil your data silently in a non-protected "pools" and "spaces" (different names MS picked up for old Logical Volume Manager concept).
Also Google MS forums for ReFS issues. There are A LOT of them. Everything looks at least half baked (not to say half ...). ReFS should be AVOIDED in a
production until at least next version of Windows or at least Service Pack.

3) MS had presented a set of interesting features (like clustered storage spaces) but they are unusable with iSCSI (or FC or FCoE or AoE etc) so far. Dedupe does
not work with hot data (VHDs) either so... Windows Server 2015

We're currently designing a new infrastructure with new StarWind SANs running on Windows Server 2012. Some questions:

• Windows Server 2012 improves network teaming. Our SAN will have six NICs with five being dedicated to the server backbone and the sixth for the general LAN. We have the option to use either iSCSI multipath or Windows Server 2012 network teaming. Both will bring NIC redundancy and spread the load. Any comments on which is preferred?
• Windows Server 2012 introduces resilient file system. The name make it sound very attractive for a SAN environment where resilience is critical. However, it's a) a new technology and b) it has some limitations (like no sparse files). Any thoughts on whether or not it's a bad idea for the host SAN drives?
• Any other Windows Server 2012 comments/suggestions/recommendations/best practises around StarWind?

Cheers, Rob.

[robnicholson]

Thanks Anton - that answers my question pretty clearly So MPIO & good-old NTFS it is.

Cheers, Rob.

[anton (staff)]

Absolutely. Good news - V8 beta with flash cache, log structure (VM-aware FS), strong digital signatures etc is coming

Thanks Anton - that answers my question pretty clearly So MPIO & good-old NTFS it is.

Cheers, Rob.

[robnicholson]

Flash cache - bit like the Drobo 1200 does? If so, cool!

Cheers, Rob.

[anton (staff)]

Drobo does wierd thing: their flash cache runs behind slow 1 GbE connection. Our flash cache will run directly on hypervisor. In a nutshell: no network I/O on cache hit.

Flash cache - bit like the Drobo 1200 does? If so, cool!

Cheers, Rob.

[Drobo]

The B1200i uses a SSD tier to dynamically boost transactional performance at the block level. It is not a cache, but an actual tier of data. Placing the cache on the SAN provides acceleration to all hypervisors without the need to purchase hardware for each physical host. Since transactional data is made up of very small blocks, it is not throughput limited and can achieve full spec on a single NIC. Should it use larger blocks, we support multi-pathing to provide higher throughput as well.

[anton (staff)]

1) Upcoming V8 from StarWind will have both flash cache and automatic storage tiering. Each approach has its own pros and cons and it's up to user to pick up the way to go. StarWind is flexible. Drobo is not.

2) StarWind runs caches (RAM and flash) directly on hypervisor so with cache hit situation request get served immediately. With Drobo even with cache hit request still has to trevel down the wire. This is SLOW. StarWind is fast. Drobo is not.

The B1200i uses a SSD tier to dynamically boost transactional performance at the block level. It is not a cache, but an actual tier of data. Placing the cache on the SAN provides acceleration to all hypervisors without the need to purchase hardware for each physical host. Since transactional data is made up of very small blocks, it is not throughput limited and can achieve full spec on a single NIC. Should it use larger blocks, we support multi-pathing to provide higher throughput as well.

[DUOK]

Dear Anton,

When, when, when?
(When V8 beta available)

[anton (staff)]

Soon, soon, soon

P.S. Don't shoot the pianist he's doing his best (c) ...

Dear Anton,

When, when, when?
(When V8 beta available)