Windows 10 Volume Replica vs StarWind
hi,
i know server 10 isn't finished yet, interested in storage replica feature believe synchronous synchronisation. wondering if else had more experience testing feature , explain of major differences between microsoft's replication , starwind replication. need hardware ms way, special? can run on 2 node cluster? if same job, reasons have purchasing third party product, can ms implementation virtual san , competing against likes starwind , datacore?
i interested know technical details of differences, , why 1 better other.
cheers
steve
hi,
i know server 10 isn't finished yet, interested in storage replica feature believe synchronous synchronisation. wondering if else had more experience testing feature , explain of major differences between microsoft's replication , starwind replication. need hardware ms way, special? can run on 2 node cluster? if same job, reasons have purchasing third party product, can ms implementation virtual san , competing against likes starwind , datacore?
i interested know technical details of differences, , why 1 better other.
cheers
steve
steve,
few disclaimers start from...
disclaimer1: work starwind software (heck, started company 1 more guy years ago , still own noticeable part of it) while i'm doing best honest cannot believe me 100% :) making long story short: never trust , double-check vendor says you. @ least we're publishing step-by-step guides or "cooking books" , you're welcomed use these logs run experiments. there's no guarantee did not work not work you. , vice versa :)
disclaimer2: windows server 10 half-baked it's not beta lot of things don't work expected, performance sucks badly affecting functionality (see more below), lot of things locked, msft not sure deliver promised ga or not. it's premature build company storage strategy on see now. lot of under nda , prefer not tell anything. i'm under nda thing can tell - stay tuned, there more interesting things coming you're going in love :)
now storage replica vs. starwind...
1) technical. storage replica mini-filter based logical volume synchronous replication, similar steeleye datakeeper , maybe double-take do. starwind virtual san name says - san firmware running on windows platform , brought 1 level run on hypervisor host, partially user-land service , partially kernel-mode driver. storage replica copies data blocks coming 1 volume volume , starwind represents distributed high performance multi-path lun. i/o "bond" local node avoid network transactions , tcp stack bypassed lower latency. lun "looks" it's iscsi , it's not (something similar smb3 smb direct connection starts life tcp , turns rdma, starwind starts tcp , dma on local node). storage replica there's component sitting above volume , "branching out" writes (if know drbd or hast - similar). "by design" storage replica , virtual san totally different.
2) positioning. storage replica disaster recovery software high uptime (nearly 0 downtime scenarios, see below). starwind business continuity solution optional dr component (async replication) 99.99% , 99.9999% uptime. while it's possible achieve of scenarios storage replica (we did sofs cluster , ha vm) it's not goal of it.
3) features. starwind "spoofing" places huge amounts of distributed write cache (ram-based) on every node, storage replica other side uses log disks (that's why ssd preferred). starwind "all-active" while sr "active-passive" design. not expect high iops sr ga code. starwind other cool things "flash-friendly" in-line 4kb deduplication (msft can 32kb block off-line 1 , steal iops storage array data written twice , 1 read required optimization process, on heavily loaded storage either have less iops or no dedupe optimization process never kicks in...). starwind can use ram not write-back cache (afaik w10 msft not read-write csv cache , cannot use non-flash write cache <-- double check this) in-memory upcoming vdi accelerator project. starwind log-structured file system there's no data , log , sr log re-read , decoded put on primary storage. making long story short: starwind kills random 4kb i/os , eliminates i/o blender effect, msft cannot (lsfs not panacea - there scenarios don't work it...). starwind offloads snapshots cheap nodes saving primary flash hot data, so-called "inter-node tiering", netapp & nimble can , msft moves hot , cold data between flash , spindle on same node only. etc etc etc
so... see sr basic important step in right direction. if find sr "good enough" - use it. if not - you're welcomed deploy third-party software. msft years leave huge holes in product strategy (i have impression quite lot of teams run engineers , not guys buying similar solutions real customers) leaving lot of space other guys. remember: companies create products , excellent companies create infrastructure :) msft company allows isvs live , grow. we're happy msft in general , windows 10 "storage" in particular.
now true value you. we've experimented run sr in different set of scenarios. some:
1) failover file server 2 nodes , no shared storage. works failover not transparent (probably because performance sucks, hope closer beta no need tune failover timeout required have 100% transparent 1 ca smb3 shares provide). please see blog:
storage replica: general-purpose file server no shared storage!!
http://www.starwindsoftware.com/blog/?p=25
2) scale-out file server 2 nodes , no shared storage. works sr "active-passive" design whole sofs idea bit compromised... see:
storage replica: scale-out file server no shared storage!!
http://www.starwindsoftware.com/blog/?p=42
3) hyper-v cluster ha vm , 2 nodes only, again no shared storage. works. can e-mail me or pm me , i'll give post draft, not published yet.
4) hyper-v cluster guest vm cluster. not work. because bugs in windows 10 guest vm cluster on normal shared storage not work well. again not published yet e-mail me.
hope helped bit :)
good luck , happy clustering!
Windows Server > File Services and Storage
Comments
Post a Comment