New switches, yes?

platypus

platypus

platypus

Caporegime
Joined
25 Jul 2003
Posts
40,311
Location
FR+UK
A small-med network of an average of 60 active network connections at any one time.

Two servers (Dual [email protected], 4GB Ram) running MSSQL Server 2000 databases, two remote servers (Single [email protected], 3GB Ram (?!)) serving 10 remote desktop sessions. Lots of workstation clients. Three 24 port 10/100 switches. Malformed packets flying everywhere between clients and servers.

Database access is being crippled at present, queries are taking ages, huge hits on the database making the system hard to use.

The obvious answer is new gigabit switches, but money is a problem..
 
and you definately know that its the network crippling your SQL performance? Might be worthwhile performance tuning your SQL and database schema before throwing money at it. Its really really easy to make SQL go slow.
 
How much have you got in the pot for switchs? Also is the network going remain flat?

Have you used wireshark to trace any network packets/frames? Have you also ran a perfmon against the SQL server?
 
Sin_Chase said:
http://forums.overclockers.co.uk/showthread.php?t=18079739

There's one if the 1800 series took your fancy. Feel free to make an offer.
Click to expand...

No offence but given that the 1810-24G's are £200ish list new with a valid warranty (as IIRC it is not transferable past the original owner) and the 1800 series frequently go for just over half that on fleabay that's a pretty poor price you're asking.

I'd echo the comments of others in this thread though that it is worth looking at the other systems at work here as a 100mbit network should be more than able to handle what you're throwing at it.

Switch wise though the entry level HP kit you can't go wrong with, further up the new 2520 series seem to give a lot of bang for buck (£1k for a 24port GigE with full PoE+ support) too and are proper fully managed devices as apposed to just web-managed.

Juniper have their entry level EX2200 line which come it at around £1k a switch and are probably pretty good, Cisco have the 2960 series at that sort of price point too.
 
My SAN is running on HP 2510 switches - been flawless so far

The only thing I dont like about them is the VLAN management terminology - it's a bit too much like Netgear for my liking in that area. Why can't all switch manufacturers deal with VLANs like Cisco?
 
As above i have used a HP ProCurve Switch 3500 for our iSCSI traffic. However, if we had the money in the pot i would have used a Cisco 3750.

Thats one of the reasons why if you require features on your switch Cisco is the only one to choose.
 
Wireshark is reporting malformed packets left right and centre, between the remote terminal servers and the two servers, nothing between the servers and the clients, but everyone is experiencing slowdown. There doesn't seem to be a performance issue with the database (local access is spot on), cache hit ratio is averaging 99% for example. Part of the problem atm is that the customer doesn't want to spend any money.
 
Last edited:
Is it possible to isolate any of the three switches to see if you have a faulty one? Seems very unlikely all three would suddendely go bad and start creating faulty packets.
 
iaind said:
The only thing I dont like about them is the VLAN management terminology - it's a bit too much like Netgear for my liking in that area. Why can't all switch manufacturers deal with VLANs like Cisco?
Click to expand...

Actually Cisco's methodology for VLAN config is damn stupid, everybody just likes the familiarity. It's far too easy to accidentally break things when adding new VLANs, the Foundry/Brocade method is far superior in many ways (one of the few things they do better) - forcing you to tag the port within the vlan config.

I'm not wild about the HP units config in general myself, good, capable switches make no mistake but I don't much like them. (The switch equivalent of a fortigate firewall actually - good but with a terrible interface)

Obviously Juniper's is, as with everything, the best going, it's just complex and very different to what everybody else does.

As to the OPs question, unless you have really terrible switches then it's something else. You really don't fancy new Gigabit switches for what you want to do, we still have maybe a hundred or so users on 100MBps switches and maybe 4% of our servers (which works out about 550 machines) with no problems at all. Throwing bandwidth at the problem isn't a good solution.
 
bigredshark said:
Actually Cisco's methodology for VLAN config is damn stupid, everybody just likes the familiarity. It's far too easy to accidentally break things when adding new VLANs, the Foundry/Brocade method is far superior in many ways (one of the few things they do better) - forcing you to tag the port within the vlan config.

I'm not wild about the HP units config in general myself, good, capable switches make no mistake but I don't much like them. (The switch equivalent of a fortigate firewall actually - good but with a terrible interface)

Obviously Juniper's is, as with everything, the best going, it's just complex and very different to what everybody else does.

As to the OPs question, unless you have really terrible switches then it's something else. You really don't fancy new Gigabit switches for what you want to do, we still have maybe a hundred or so users on 100MBps switches and maybe 4% of our servers (which works out about 550 machines) with no problems at all. Throwing bandwidth at the problem isn't a good solution.
Click to expand...

You're right - it is familiarity - it's what I started to get involved with VLANs on so it's what makes sense to me

Netgear are appaling in that regard, you can have 2 switches with identical config which will behave completely differently because they have slightly different firmware versions!
 
iaind said:
You're right - it is familiarity - it's what I started to get involved with VLANs on so it's what makes sense to me

Netgear are appaling in that regard, you can have 2 switches with identical config which will behave completely differently because they have slightly different firmware versions!
Click to expand...

Cisco don't get off without criticism in that regard though, last time I had cause to configure it jumbo frames was a system parameter rather than a configuration one. So copy the config to a new unit and you don't have jumbo frames any more...
 
platypus said:
So what is?
Click to expand...

Well it depends (unhelpfully) what the problem is. I'd be trying to tie down the problem to an individual switch or port, it sounds like a potential duplex mismatch or similar as a starting point.

I'm fairly sure replacing everything would fix it but it's a little drastic and there is the potential to be left with the same problem. There not much chance a switch itself is mangling packets (switches are too dumb usually, takes a router to properly mangle traffic) so it's most likely a link somewhere...
 
You must log in or register to reply here.
Back
Top Bottom