Newbie datacenter lesson #2: Performance considerations
So you are finally doing it! Moving everything to a colo site - so exciting! So much redundant input power! So much reliable A/C!
So much latency between sites.
It made so much sense to move EVERYTHING over that the only consideration you gave any thought to was DFS/Folder Redirection (which does, indeed, stink over the WAN). One other small fuzzy area you forgot about was AccPac (Sage ERP 200).
DFS/Folder Redirection - at least this was a known potential issue. It's slow, but users can live with it if they must (and $ dictates wait they must). But AccPac! It's so un-useable at this point that immediate measures must be taken.
Latency, you ask - how awful is it? 27ms. (both sites on 100mb fiber)
Yep, 27ms is enough to bring AccPac to its knees. It turns out that AccPac is a rather old application, and has a neat feature called 'DBspy' that displays its age nicely. If you watch the 'DBspy' monitor screen, simply clicking on a single widget-thing in AccPac yields dozens of SQL calls. It's no wonder the entire app is slow now! Going from <1ms to 27ms is a fairly significant multiplier.
Speaking with our AccPac rep, he informed us that what we are trying to do (workstations connecting to remote SQL) is not a good idea, and we have two alternatives:
So much latency between sites.
It made so much sense to move EVERYTHING over that the only consideration you gave any thought to was DFS/Folder Redirection (which does, indeed, stink over the WAN). One other small fuzzy area you forgot about was AccPac (Sage ERP 200).
DFS/Folder Redirection - at least this was a known potential issue. It's slow, but users can live with it if they must (and $ dictates wait they must). But AccPac! It's so un-useable at this point that immediate measures must be taken.
Latency, you ask - how awful is it? 27ms. (both sites on 100mb fiber)
Yep, 27ms is enough to bring AccPac to its knees. It turns out that AccPac is a rather old application, and has a neat feature called 'DBspy' that displays its age nicely. If you watch the 'DBspy' monitor screen, simply clicking on a single widget-thing in AccPac yields dozens of SQL calls. It's no wonder the entire app is slow now! Going from <1ms to 27ms is a fairly significant multiplier.
Speaking with our AccPac rep, he informed us that what we are trying to do (workstations connecting to remote SQL) is not a good idea, and we have two alternatives:
- Move the server back.
- Set up terminal services/appServer at the colo, have the users connect via RDP/remoteApp instead.
I don't really want to mess with the finance folks' workflow, so local server it is. Thankfully we only have 15-20GB of databases to move. Does mean one more SQL Standard license to purchase, however. Good thing we left that single Hyper-V box behind for domain controllers...
So the lesson is: Check with ALL of your vendors and review the planned changes with them, especially the legacy providers.
Comments
Post a Comment