Write-back or write-through in Starwind

[mkaishar]

I am little bit confused about which to use within Starwind Targets

Storage server has 24GB soon to be 32GB
R710 with MD1220, H800 with 1GB Cache

MPIO/multipath is configured and working successfully for both vmware and windows targets

I enabled 1GB write-back cache for each vmware target (we have 12 targets each is 750MB in size)
I enabled 1GB write-back cache for each windows target (SQL, Exchange, applications, etc...)

We do not have HA

So is it better to use write-back or write-through in Starwind?

Is there a white-paper on this somewhere on the site?

Thanks,
Mark

[anton (staff)]

With HA you can use Write-Back Cache for any scenario. For non-HA you have to stick with Write-Thru Cache unless your scenario allows target to fail (say it's Tier2 storage or maybe destination to overnight back up).

[mkaishar]

Any white-papers on this?

Technical reasons against wb in non-ha environment?

thanks,
mark

[Max (staff)]

No whitepapers on this, everything's simple:
With write through your data is written simultaneously both to disk and to RAM, thus preventing your system from a data loss in case of a power outage or malfunction. However you sacrifice the array performance.
Write back caching mechanism assumes writing the data to the RAM, actual write to disk will be started when background lazy writer decides it should flush cache line to disk. This mechanism is A LOT faster but it is highly power-dependent (that's why most of the RAID controllers have a backup battery.) As your machine is not dedicated hardware and has TONS of software and hardware you can have crash or hang or BSOD virtually any time (so logically replacing battery with UPS does not work). This means many gigabytes of unsaved writes will be lost. Once and forever. HA has distributed cache so no way for all nodes go down at the same time.

[mkaishar]

I understand the concepts of caching, I just thought Starwind's implementation was magic and something totally different

[anton (staff)]

I'm not so sure... StarWind has *distributed* Write-Back Cache. This means you can have multi-gigabyte Write-Back Cache enabled (with HA configuration of course) for say SQL Server database transactions log. No single instance hardware RAID controller can do the same.

I understand the concepts of caching, I just thought Starwind's implementation was magic and something totally different